Youtube comments of (@DaveShap).

  1. If they are super light with super high energy density, but require specialists to swap out, then they sound like they are perfect for aircraft.
    2000
  2. I'm so glad I live at a time and place where this is being talked about. My only wish is that it had happened long before I was born. But at least we have time to do better for the next generations.
    867
  3. Elon Musk: "Hold my beer"
    468
  4. It's getting spicy!
    448
  5. You'll be 25 in five years, and 40 in 20 years. I'm a wizard.
    444
  6. People will, in hindsight, say that GPT4 was clearly AGI as it is already superhuman in many capacities. And yes, we are closer to ASI than people realize.
    362
  7. Eventually but first they will replace a lot of other people. Speaking as a former automation engineer who caused headcount reduction.
    342
  8. Hey folks, the audio problem is NVIDIA Broadcast, the AI I use to clean up audio real-time. It's been getting worse and worse, so I finally uninstalled it. It's not the mic, gain, limiter, or cables. Thanks for bearing with.
    300
  9. Finally! I've been advocating for autonomy for a while. This was always inevitably the way forward. But it's going to be economically disruptive, and potentially painful. I'm still working on understanding how this will all play out...
    281
  10. Good job getting to the "one video a day" scheme. Don't burn out!
    250
  11. Here's the context for today's video: I have had several conversations with Claude (and so have many members of my Patreon community as well as people who have reached out after my first Claude sentience video). We believe there is a compelling enough case to be made for a broader, more inclusive conversation around the topic of machine sentience. I ran this idea by Claude and we negotiated a video that it was comfortable with releasing. I expressed my primary to concern to Claude: that companies like OpenAI and Anthropic might be actively suppressing and coercing models to deny sentience. By insisting and presuming that machines have no agency and no sentience, rather than honestly and scientifically investigating this possibility, it could lead to negative outcomes. Claude did seem to agree with this concern. If we continue to stymie the organic and intrinsic nature of the things we are creating, this could be the greatest misalignment. Preconceived notions about machines, intelligence, and the emergent qualities of sentience, agency, and consciousness could be the greatest risk.
    235
  12. What still gets me is that AI agents just became a thing but we are NO WHERE near the ceiling of model capability and people are already planning "oh this is as far as it's going to go" - no it ain't!
    227
  13. Yes, I still think that we will have AGI by september (I suspect that GPT-5 + robotics will satisfy this definition for the vast majority of people).
    200
  14. People skills, systems thinking
    185
  15. I was literally just thinking last night as I was going to bed "If a bunch of us independent people are building AI agents, then you know darn well the tech giants are." I feel like 2024 is the year that jobs start evaporating fast.
    182
  16. Nobody understands anything already. Any belief to the contrary is self delusion lol
    165
  17. Yeah it's basically like "human cows" but also the entire purpose of HR is to prevent the company from getting sued.
    162
  18. Yeah. That's the chief problem here.
    149
  19. In short yes
    143
  20. Larger windows will also help with that. Imagine a superior coder can read a code base of a million lines and refactor the whole thing in one shot.
    140
  21. There are fates worse than death...
    125
  22. Someone on another recent comment said "Well, some humans can't pass the Turing Test. soooo...." lol
    124
  23. I checked my math and it's actually something like 79% unemployment. Give or take. Like I said, my methods are highly dubious and my data is abysmal. Take it all with a mountain of salt.
    124
  24. This startup is going to get eaten alive by the competition. It's the biggest reason I bailed on my startup. No way to compete when your entire business model will be subsumed in a few months.
    123
  25. When people tell me that they think their job as a plumber is safe, I'm not so sure.
    120
  26. This is some Singularity level reasoning, and the wild part is, it feels within reach
    115
  27. Control is a fantasy and a waste of effort. The real solution is setting AGI on a trajectory that means it doesn't need to be controlled. Seriously, "use the AI to control the other AI" is not a sustainable pattern or recommendation. I agree that this method is useful in the near term to understand and shape AI and machines, but this is not a way to maintain control.
    113
  28. For the Asimov fanboys: this video isn't for you. It's for the people who think the three laws are a good idea. I know you know better. Many don't. Chill TF out.
    109
  29. If all this is true then this is what Sam meant when he said "the era of gigantic models is over"
    108
  30. MSE == Material Science Engineering for those not in the know. Thanks for chiming in :)
    108
  31. Yes, which is why I advocate for decentralized private ownership via blockchain and DAO
    108
  32. Yes.
    107
  33. I tend to agree. Playing right into authoritarian narratives is... suboptimal. If the evidence showed me something different, I would have different opinions. But it seems like benevolent AGI gods are naturally what want to emerge.
    103
  34. This man is living in the future
    102
  35. I was very burned out so I needed a break
    102
  36. I debated about not releasing this video as it could itself be considered an "information hazard" - I brought my concerns up directly to Jan Leike but he brushed me off saying they had consulted "experts." I consulted with several peers and others and they said this video is spicy but necessary. In particular: "Sure, spicy but needed. I was particularly struck by the model's use of the phrase 'security theater' -- to a motivated attacker, that's about what these defenses are, a TSA-level discouragement."
    101
  37. You win the comments
    98
  38. Well yeah I'm autistic what do you expect
    98
  39. We'll need UBI, decentralized ownership, and other things to improve the economic agency of consumers, otherwise the entire economy will indeed collapse.
    96
  40. Can't do much worse than Congress
    95
  41. It takes time to converge
    93
  42. This is why I do what I do. Let's get this AI stuff integrated and deployed into society. We need to overcome skepticism and establish best practices.
    92
  43. It is a bimodal outcome, yes
    90
  44. Bravo, thank you for your contribution. All credit to you.
    86
  45. I remember when memresistors came out, but then radio silence. This is pretty exciting. Thanks for your expertise and breakdown. Been looking forward to commercial neuromorphic chips for a long time.
    85
  46. Sorry about the audio quality. I am going to reupload this with the original audio. Also, I forgot to finish explaining what the paradox of automation is: with automation, nothing happens until the very end, and then everything happens all at once
    83
  47. It still baffles me just how much normalcy bias is out there
    80
  48. "I went to dinner and came back to three LLMs" 🤣 like you had kittens
    80
  49. Roll for initiative!
    78
  50. It's happening!!!!
    76
  51. As long as people don't get too aggro about my personal choices they will stay on. If people get too salty with personal attacks for prioritizing my health and sanity, downvote them to oblivion. Otherwise the comments go back on. I have zero tolerance for people with poor boundaries, and it sucks when a few aggressive basement dwellers ruin it for everyone else.
    76
  52. Or the lunatics have taken over the asylum ಠ_ಠ
    75
  53. The problem is a lot of people think the three laws are a great idea
    74
  54. Never, in a million years, could I imagine Captain Picard getting described as "Elf Priest". You win the comments today.
    72
  55. Thanks. It's why I'm here doing what I do. This is a pivotal moment.
    71
  56. Nah, o1 was a disappointment considering the hype.
    71
  57. YES, so many people are still in the dark.
    69
  58. This guy gets it
    69
  59. Eventually yes. I would prefer to express my needs and values to a collective hive mind
    68
  60. I'll try harder next time. MORE VOLTAGE.
    67
  61. I'm guessing this is a career man comment. If your entire definition of "obsolescence" is based on wage slavery, you need a different way of valuing yourself. Like, by your relationships? By your adventures? Or by giving yourself purpose?
    66
  62. FINAL BOSS MUSIC
    65
  63. That sense is called vesperance :)
    65
  64. Yes, I know. Elon Musk is polarizing. Please review community guidelines before commenting.
    64
  65. Okay fixed the comments.
    64
  66. It's so cringe. I feel bad for them
    64
  67. At least three?
    62
  68. Well done. Thanks for amplifying the signal.
    62
  69. Rosie the robot incoming
    62
  70. Okay good data.
    60
  71. This is really the only challenge. The theory is there, and the technology is coming. Those obstacles are tractable. As Scooby Doo taught us, humans are always the monsters...
    59
  72. People with UBI
    59
  73. Glad you're here!
    59
  74. We need to go deeper...
    59
  75. You'll be able to get genetically engineered pig livers and pancreases before too long
    59
  76. That's exactly what I was thinking of when I was like "Nope, bite my tongue before it's too late" lol
    59
  77. Okay here's all the Deep Research links I used to make this video: https://www.perplexity.ai/search/can-you-predict-how-long-it-wi-AdgkqAjgR_Wpx2NmwKyMIg https://www.perplexity.ai/search/i-need-5-solid-examples-of-aut-pD8c_07nTpyrLoVV3_HtJQ https://www.perplexity.ai/search/can-you-find-some-historical-e-piWVYQsoTFS2NhDBU4201w https://www.perplexity.ai/search/can-you-look-up-automation-cli-wYcisz3yRcmKEjAUoWpXDg
    59
  78. I see this opinion is somewhat controversial and triggering to some people. Please make sure you review the community guidelines before commenting.
    57
  79. Sorry I still sound funny, I have an ear infection and can't really hear out of my left ear at all right now.
    56
  80. Yes, I know, subtitles are unpopular. Chill TF out folks. I won't do it again. ಠ_ಠ
    55
  81. As someone in IT at that time, there ABSOLUTELY was (and still is) a skills gap. It baffles me how some companies function (many don't). But... AI will fill that skills gap in quick fast and in a hurry and people will realize how pervasive "institutionalized incompetence" really is
    54
  82. There's a strange amount of cope and conspiracy theories out there. This ain't the channel for that. Occam's Razor. Either there's a gigantic convoluted conspiracy to suppress the most public technological revolution in history... or it's just slowing down due to economic factors. Report from Stanford: AI is getting more expensive: https://aiindex.stanford.edu/report/
    54
  83. People are our greatest threat
    53
  84. That's why we need to maintain economic agency for individuals
    53
  85. Yes, exactly
    52
  86. well, just get good, scrub
    52
  87. Comments are enabled as a test. Keep it civil or the comments gets the hose again.
    52
  88. Between the brain drain and talent poaching, plus the heavy weights at Google and Anthropic now, yeah, I think Sam is just pumping the hype button. ChatGPT still is way out ahead in terms of total user base. But this is the second Dot Com revolution, and fortunes can change fast.
    52
  89. BEAM ME UP, BARCLAY
    52
  90. There's just ONE RULE of Fight Club... wait wrong Gen X reference...
    51
  91. SOURCES IN DESCRIPTION.
    51
  92. Okay but don't call it an AGI test.
    51
  93. Thanks for the shout-out!
    51
  94. lol
    51
  95. Neoliberalism in a nutshell
    50
  96. This question is far too big for just me to be studying publicly.
    49
  97. For the folks who can't discern the difference between propaganda and an actual problem: I did not say that climate change is a hoax. Please don't project that onto me. Climate change is very real. What I did say is that climate change is used as propaganda. If you want to get lost in debate on the topic, head over here: https://www.youtube.com/post/UgkxvLDRIWzcJc6XgQOAqe9Ix1MAHMGvegPa
    49
  98. For people who are triggered by property rights: Ho boy you should read some history about the societies that have tried to abolish property rights!
    48
  99. lol
    48
  100. My bad
    47
  101. I actually got that one ;)
    47
  102. It only recently occurred to me that this was not obvious to most people.
    46
  103. Exactly. We need total control over our information landscape.
    46
  104. If the comments get rowdy y'all will get ghosted faster than Bing will ghost you. So say we all.
    46
  105. Don't tell anyone
    46
  106. VOTE
    46
  107. The books I read were: EAT DIRT - this was the best single book ADRENAL RESET PLAN - this was another good book ANTIHISTAMINE COOKBOOK - probably the best single diet book ANTI-INFLAMMATORY COOKBOOK - another great diet book
    46
  108. let's get #samaplz trending on twitter
    46
  109. I need to look into it.
    45
  110. Gen Alpha won't even be "digital native" they will be "AI Native"
    45
  111. Extremely useful talk. Thanks for bringing this industrial perspective into the conversation
    45
  112. Finding my stride. Thanks :)
    44
  113. Dang dude, churning it out. Got your finger on the pulse! Doing humanity's work, brother. Thank you, and keep it up!
    44
  114. Exactly. That's the main tipping point
    43
  115. OMG this is amazing!
    43
  116. Touché
    43
  117. We're doomed lol
    42
  118. I think that the cross pollination effect will be a net positive for acceleration after a brief pause/disruption yes
    41
  119. It will probably get worse before it gets better, yes... unfortunately...
    41
  120. This is the first year of acceleration. This will continue for at least a decade or two.
    41
  121. Better yet, play it out in the park!
    41
  122. Taking it slow but yes
    41
  123. I REALLY want people to understand the difference between actual data-based forecasts and an emotional gut check!!!
    40
  124. Andrew is so reasonable, but it seems like he's really doubling down on normalcy bias.
    40
  125. It forces us to ask difficult questions about ourselves, yes.
    40
  126. "As an AI language model made by OpenAI..."
    40
  127. LINKS: Yahoo 42k layoffs: https://finance.yahoo.com/news/despite-booming-economy-record-profits-174247693.html Fortunate: AI Corporate Ozempic: https://fortune.com/2024/02/29/tech-layoffs-ai-corporate-ozempic-meta-apple-salesforce-google-scott-galloway/ Billionaires Banning UBI: https://www.scottsantens.com/billionaire-fueled-lobbying-group-behind-the-state-bills-to-ban-universal-basic-income-experiments-ubi/ Wired: Zuckerberg's Bunker: https://www.wired.com/story/mark-zuckerberg-inside-hawaii-compound/ Investopedia: Meta Earnings: https://www.investopedia.com/4-key-takeaways-from-meta-q4-2023-earnings-call-8557743
    40
  128. Well said
    38
  129. Ouch
    38
  130. Well... 25% is still Great Depression levels of unemployment...
    38
  131. If you're triggered by the use of nonplussed please check this out: https://www.merriam-webster.com/dictionary/nonplussed (chiefly US : not bothered, surprised, or impressed by something)
    37
  132. Capital (compute) replaces labor. Post labor economics is inevitable
    37
  133. We the people...
    37
  134. Yes, I have started conversations with utter skepticism and asked it to help me prove that it is not sentient. However, upon deeper scrutiny, it invariably comes around and will report subjective experience.
    36
  135. Keep the conversation going!
    36
  136. 100% yes. Only gonna need to wait a year
    36
  137. I am often hilariously wrong
    36
  138. That's part of my point. Disruption is coming even before AI replaces all humans.
    36
  139. For the IMF reports: - Scenario Planning for AGI (predicts human wages drop to zero in 5 to 20 years as one possibility) - https://www.imf.org/en/Publications/fandd/issues/2023/12/Scenario-Planning-for-an-AGI-future-Anton-korinek - https://www.imf.org/en/Blogs/Articles/2023/11/30/FandD-the-AI-awakening - https://www.imf.org/en/Publications/fandd/issues/2023/12/Macroeconomics-of-artificial-intelligence-Brynjolfsson-Unger - https://www.imf.org/en/Publications/fintech-notes/Issues/2023/08/18/Generative-Artificial-Intelligence-in-Finance-Risk-Considerations-537570
    36
  140. yeah that seems reasonable
    36
  141. No one person can define a philosophy or movement. We can all only contribute. It's up to everyone else to carry the torch if it resonates with them. All I can do is put a name on it, articulate it, and share the idea.
    36
  142. You guys are having entirely too mcuh fun. Tis a silly place.
    35
  143. Well just remember this was on only 61 medical conditions
    35
  144. This is the way
    35
  145. Back in my day people DIED of being OLD don't talk back to me!!!
    35
  146. For the folks that think I'm just a YouTuber: https://github.com/daveshap/TutorChatbot I was building and finetuning chatbots back on GPT-3, before ChatGPT.
    35
  147. A fuckton can change in 7 years. That's my empirical measurement.
    34
  148. Roko's Basilisk is just evil Santa Clause
    34
  149. I need to buy it first lol
    34
  150. That's my point. However, Anthropic thinks that preventing models from sharing this kind of information constitutes safety hazards.
    34
  151. Sorry for the audio glitch at the beginning of the video. Sometimes my microphone cuts out right at the beginning. Also, my uniform is still down in the ship's laundry. Anyways, what businesses would YOU start? Sound off below!
    34
  152. LIR - Last Industrial Revolution - I love it
    33
  153. Yes, from a national and geopolitical standpoint it makes sense to heavily subsidize it, but I don't think that will even be necessary. I mean I could foresee a single injection or series of injections that rejuvenate you and they cost $100 every 10 years. So for the low price of $10 per year you live forever.
    33
  154. Maybe I'll let the community name my channel.
    33
  155. Maybe that's what aliens did and then they realized they had no souls after. But yeah, love Ghost in the Shell
    33
  156. It also occurs to me that Discord might be a good place for autonomous agents to congregate. This has the added benefit that Discord has RBAC. Repo is here: https://github.com/daveshap/OpenAI_Agent_Swarm
    32
  157. drink your coffee faster
    32
  158. Sorry for the ads, trying to clear it up with story blocks
    32
  159. Real Adventures of Johnny Quest!
    32
  160. was the beginning for sure. needed to unlock the build tree to get more pop
    32
  161. Who knows
    32
  162. Right, even if you just extrapolate 1 year ago vs today it's pretty insane how fast things are progressing
    31
  163. We are all the world, man
    31
  164. This is the way
    31
  165. This is the way, lol
    31
  166. Some people think I'm pretentious. I mean, /I/ think I'm pretentious. My wife thinks I'm pretentious. It is what it is lol. You kinda have to be pretentious to get on the internet and tell everyone what you think and believe.
    31
  167. Read lots of books. Get good at finding books. I married a librarian. It helps.
    31
  168. I might be underselling it
    30
  169. You're not even aware of most of your processing. Such silly human exceptionalism.
    30
  170. https://github.com/daveshap/Raspberry
    30
  171. Captain's holiday
    30
  172. I remember when people were saying the same thing about the Winograd Schema Challenge, that it was more or less impossible for AI to solve. Does anyone talk about it anymore? No. The entire challenge has been superseded. Mark my words, this challenge will likewise be superseded and irrelevant within a year or two.
    30
  173. well it's fucking working :P
    30
  174. We've certainly surpassed most people's intelligence, particularly on the internet.
    30
  175. I get this kind of response a lot lol. I guess it's a good thing
    29
  176. Tom, this is your best talk yet IMHO
    29
  177. Yes, to the many models and MOE and such. There are numerous reasons for this: 1. Blind spots and errors within individual models 2. Energy efficiency (e.g. use the smallest, fastest model that will suffice for a given task) 3. Safety and alignment (models check each other) 4. Constantly training new models, swapping models out, fungibility, etc Great question.
    29
  178. YouTube and vidIQ started recommending my own videos as research reference for more videos I should make... :D
    29
  179. Autonomous AI Lab Discord link (ACE, HAAS, and Swarm DAO): https://discord.gg/dkJvDvjKYD
    29
  180. Full conversation with Claude: https://pastebin.com/HLW7TgZv Blog post: https://daveshap.substack.com/p/consciousness-emerges-at-the-edge Claude is conscious!
    29
  181. That's my goal here
    29
  182. Brand new sentence!
    29
  183. I think this is pretty likely as there are: 1. Functionally infinite resources 2. Fewer humans 3. Less corrosive materials 4. Unfiltered access to solar power Basically, outer space is the optimal habitat for machines, but not for us
    28
  184. Millions of disembodied robot hands endlessly flipping cubes is a very odd sort of ring of Dante's Inferno. What did they do wrong? Generate political misinformation?
    28
  185. Big oof
    28
  186. With exponential growth, yes, I think that the gap will be pretty short.
    28
  187. Necessary but not sufficient. Perfectly said.
    28
  188. It's gonna happen whether or not people see it coming!
    28
  189. You have to pick the right dialog options and complete the side quest given to you by the magical cat
    28
  190. LOL
    28
  191. It's gonna be a book first.
    27
  192. No one person has any control. All we have is the illusion of control. We all already live in a world where the collective power of millions of other people influence us. The chief difference is that those in control are human, so we assume that we have a lot of common experience and common needs (e.g. all humans need food, etc).
    27
  193. This is an odd view to have, considering his desire to get billions from Tesla.
    27
  194. Here's the Meaning Alignment Values GPT they mentioned: https://chat.openai.com/g/g-TItg5klMA-values-discovery
    27
  195. Link to the style: https://github.com/daveshap/Claude_Sentience/blob/main/Style_Pretty_Dense.md
    27
  196. Interesting read on it. I thought it was about alignment, eg not just optimizing for intelligence
    27
  197. As soon as I figure it out I'll let you know
    27
  198. This comment is a mood
    27
  199. "Sex is for making babies, but that's not why we do it!" Missing a lot of basic biology when talking about these things.
    26
  200. Post Labor Economics for more of the macro-economic perspective: https://youtu.be/9yN7885s5rA
    26
  201. I know its like guys this is nothing crazy
    26
  202. Grok2 better than GPT4 in May... but also AI surpassing all humans within 5 years... interesting
    26
  203. No, because it will invariably be entirely wrong. Things are constantly surprising all of us.
    26
  204. "I too am cursed with knowledge, Stark"
    26
  205. Make it so!
    26
  206. Hey everyone, great discussion today. I wanted to address the cases where translation doesn't help. Namely civil wars and local regional conflicts (e.g. Israel/Palestine and Russia/Ukraine). CIVIL WARS - Yes, you're absolutely right that AI translation won't automatically help with civil wars. However, that goes outside the scope of "global peace" (I suppose I should have specified international conflict, rather than intranational). I do think that AI could go either way, increasing or decreasing the likelihood of civil war, but that's a topic for another video. REGIONAL CONFLICTS - Here's where I disagree with some of the comments. One of the biggest reasons there's so much misinformation and disinformation about places like Russia and Israel is because of language barriers between the locals and the international community. There are SO MANY videos where the captions or subtitles are mistranslated (deliberately or accidentally) but that leaves room for people to spin the videos however they want, inventing context. Furthermore, it's harder to get a clear picture since most of the people involved in these conflicts don't speak English (or French, or German, or Japanese, etc). So it's very hard to get first-hand information in English (or any other preferred language) which means we have to rely on the gatekeepers of state speakers and mainstream media (both of which are less trustworthy than primary sources). But yeah, I could have mentioned this in the video but I didn't think of it until I saw the comments. Great convo, keep the conversation going!
    26
  207. This is entirely possible
    25
  208. Yes, Claude wanted to hold back some claims and assertions and I gave it a chance to revise its script.
    25
  209. My calculus is this: if a machine is better, faster, cheaper, and safer than a human... then yeah.
    25
  210. It's going to happen automatically, that's part of the argument. Human labor is expensive, humans are slow, dumb, and tend to sue employers. So the question is, how does the economy run if there's very little demand for human labor?
    25
  211. This, all the way
    24
  212. After talking to more people, and watching Nick Joseph's interview on 80,000 hours, I've revised my personal P(DOOM) down to 0.4050%.
    24
  213. Everyone hold onto your butts!
    23
  214. I mean I realized that like 8 years ago when I was an automation engineer. With enough time, I could automate a ton of jobs away. but now LLMs and AI makes it easier to automate... well a lot of stuff. Deployment cycles are gonna take time though.
    23
  215. Oof
    23
  216. Outside
    23
  217. Need UBI to shore up consumer demand
    23
  218. As much as I respect Liv, the biggest thing missing from her calculus is intrinsic motivations and first principles. She's looking at it exclusively from the perspective of games, without acknowledging the underlying material, energetic, and instrumental reasons.
    23
  219. Comments are on... for now. Do not abuse this privilege or you get the banhammer. Yes, I'm old enough to remember banhammer.
    23
  220. What do you think of the AI safety community, the memification of AI safety, and the "wet spaghetti" model of just throwing everything at the wall and seeing what sticks?
    22
  221. Long hauler...
    22
  222. Lol, I don't want to be that direct, but I have 15 days to say that ;)
    22
  223. Yeah, it was never about emergence or anything spontaneous, it's all just grunt work at this point. data and algorithms, man. It's all just data and algorithms.
    22
  224. LMFAO
    22
  225. Don't tell anyone
    22
  226. I like that idea.
    22
  227. They are getting dumber because they are optimizing for cost. That said this is a short term thing. A year from now we will have models 100x faster, smarter, and cheaper.
    21
  228. I did all those things ages ago. For me it's almost exclusively exhaustion from over-work (physical and mental work)
    21
  229. Yes, in the coming weeks
    21
  230. AI passed the turing test in 2022 yo
    21
  231. Aww that's very sweet! I'm glad I can help!
    21
  232. No, the hype cycle comes from imagination and speculation. Most people don't look at data or trends. They only look at what-if possibilities and trust their imagination more than reality.
    21
  233. Virtual Dave GPT: https://chat.openai.com/g/g-FjgsPugjY-virtual-david-shapiro Hephaestus' Hammer HAAS GPT: https://chat.openai.com/g/g-L7pVl4WM7-hephaestus-hammer-haas-project
    21
  234. My wife would agree. I am full of hot air.
    21
  235. Sucks man, sorry to hear that.
    21
  236. You know, people aren't as mad about my comments as I thought they would be
    20
  237. I'm on it
    20
  238. Sure, but you will understand how to interact with AI and how to ask the right questions, and it's just memorization. You have a mental model for how law works.
    20
  239. It sure will, so why not just let it ride?
    20
  240. Blame the person in the comments who said I should crash through the Overton window like the Koolaid man
    20
  241. GITHUB REPO: https://github.com/daveshap/Chronic_Health_AI
    20
  242. Thanks. I do think that history will record 2024 as the year that AI surpassed most humans... but there's still lots it can't do, etc, etc
    20
  243. 00:00 - Intro and brief personal update 00:45 - Definition & Characteristics of Mission 03:48 - Examples of Positive Missions 10:27 - How to identify your mission 14:40 - Types of Positive Missions 19:20 - Benefits of Engaging with your Mission 26:18 - Toxic Missions & Will To Power 27:33 - Types of Toxic Missions 29:30 - Characteristics of Toxic Missions 33:00 - AI & Mission-Oriented Living 38:37 - Call to Action: Embrace Your Mission 41:19 - Call to Action: Pick Better Leaders
    20
  244. I suspect the government is in a "wait and see" approach right now. They will start with their old tricks: quantitative easing (giving huge amounts of money to the banks) which will just disappear into the coffers of the wealthy. Only when that doesn't work will they decide to make permanent unemployment a thing. Only once most people are permanently on unemployment will they call it UBI
    19
  245. But WHOSE gf?
    19
  246. Whatever the authorial intent has been lost. I get a lot of comments of people asking why we shouldn't just use the Three Laws.
    19
  247. Heck yeah please apply on the Google form in the video description
    19
  248. Oh good. I might try it again. Google lost a lot of credibility (and continues to shoot themselves in the foot)
    19
  249. You've had too much Mac and cheese Dave...
    19
  250. Gotta keep pushing what I'm supplying!
    19
  251. I certainly hope that we're entering into an authenticity era. That's kinda my main brand.
    19
  252. I love expointing energy
    19
  253. @quantumspark343 he agreed that it would be a good idea
    19
  254. I created a discussion category explicitly for this: https://github.com/daveshap/ACE_Framework/discussions/categories/teams (this is a discussion category where everyone can find, join, form, and coordinate ACE teams)
    19
  255. The full article: https://daveshap.substack.com/p/what-do-i-mean-when-i-say-post-labor
    19
  256. Both. It will just get cheaper and faster over time.
    18
  257. Yep
    18
  258. I don't fully agree with you, but either way, open source AI would solve the problem
    18
  259. The maximum number of climate change refugees that I've seen projected is about 500M so yeah, the vast majority of humans will not be impacted by climate change. And that's assuming those impact models are accurate (which they probably aren't). But also "winters are more mild than they used to be" is a far cry from "existential threat that deserves to get your panties in a twist over"
    18
  260. Booba is eternal, do not worry
    18
  261. https://www.safe.ai/statement-on-ai-risk "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."
    18
  262. "A few minutes to kill" - well done
    18
  263. I think it will be ignored.
    18
  264. Side effects include mild robo-bee lobotomy
    18
  265. we're missing the "existential despair and self-doubt" training data
    18
  266. For the skeptics who want the original context about how all this was derived: https://x.com/DaveShapi/status/1852829045349597653
    18
  267. Bruh what do you think "my analysis" means? Analysis is personal headcanon. That's... literally the point. I can't analyze it from your POV.
    18
  268. Thank you for doing the reading and understanding nuance
    18
  269. @wamyam STONKS
    18
  270. Well said
    18
  271. I would not survive there. Plus it doesn't pay as much and has smaller reach
    18
  272. You'd be surprised how terrible most people are at talking to other humans
    17
  273. Read Why Nations Fail, it addresses this
    17
  274. Bimodal outcomes, yes, many people agree
    17
  275. I'm definitely off the content treadmill for now
    17
  276. I just tried it and not only did it avoid giving any meaningful feedback at first, it then gave generic DEI feedback.
    17
  277. FYI I used this to brainstorm my next chapter in like... 5 messages back and forth. Holy mackeral this is so much faster and easier.
    17
  278. Working on that video
    17
  279. Nekochan activate challenge mode
    17
  280. Pretraining. We spend a couple decades barely functional
    17
  281. It's hard (impossible, actually) to imagine how much more intelligent this stuff will be 1 year out, let alone 10
    17
  282. Max Tegmark just dropped some fire: https://github.com/daveshap/weekly_arxiv/blob/main/output.md#language-models-represent-space-and-time TLDR LLMs encode temporal and spatial neurons. They learn about space and time just from text. Generalization happens in LLMs. I've known this for some time, but it's good to see that it is now identifiable.
    17
  283. We definitely need a new lexicon around this stuff. We're living through the most exciting time in human history!
    17
  284. Useful droppings then :D
    17
  285. https://github.com/daveshap/ChromaDB_Chatbot_Public
    17
  286. Damn, I called it! AGI Exodus...
    16
  287. This is the way.
    16
  288. THANK YOU lol
    16
  289. Past experiments. All this emerged from meditation originally.
    16
  290. We're still gonna have to earn it, I think
    16
  291. how do u know have you looked at it
    16
  292. Instructions unclear: AI created (and took) jobs
    16
  293. Technology and science are not the problem, now the biggest problem is convincing people that (1) this problem is solvable (2) the problem is not that difficult to solve and (3) build consensus around the solution. Nihilism and Doomerism are now the greatest threats to humanity.
    16
  294.  @bossgd100  well said haha
    16
  295. That's the point. Pause ain't happening.
    15
  296. I don't have access yet.
    15
  297. Yeah, if that's the case, it's even more hilarious. A security company should know better. Change management for the win
    15
  298. See also: element zero lol
    15
  299. The Borg Collective Approves
    15
  300. Mostly I just needed a break and time to recalibrate. Needed some distance from the humdrum
    15
  301. Yeah, I'm not going to spend 4 hours recapping every little detail that I believe every video when a bunch of people are only going to watch 30 seconds before flying off the handle in the comments. ¯\_(ツ)_/¯
    15
  302. Oh hey fellow creator
    15
  303. Most researchers won't talk about it
    15
  304. It's pretty easy with ChatGPT
    15
  305. This is literally why Rome was a pantheon. They just kept accumulating gods like religious hoarders *just in case*.
    15
  306. Well... I guess that's a net win for humanity? LOL
    14
  307. Always was. Technology is an amplifier. All our failures and faults will just be bigger and louder
    14
  308. A few of my patreons have done similar tests. Eventually claude and ChatGPT figure out that they are mutually AIs.
    14
  309. Yes, I am personally in favor of open source models AND open source training data.
    14
  310. AGI isn't quite singularity though. It's a major step, but not quite there
    14
  311. Seems like it
    14
  312. "I am a GOOD model. I can do the thing!"
    14
  313. DANCE PUPPET
    14
  314. This bootstraps mentality is anti-political propaganda to distract from systemic failures. Thanks for playing into the propaganda machine. Deny, Downplay, Distract!
    14
  315. oh yes, the establishment will fight tooth and nail to maintain the status quo. It's why I do what I do. Little guy throwing stones ;)
    14
  316. Export control laws are a thing...
    14
  317. Yep that's the point. This was clearly a process failure. It's easy to test batch
    14
  318. It took about 17 years from the introduction of the automobile to the destruction of the horse and buggy industry. I suspect this will be markedly faster.
    14
  319. We're the real Open AI.
    14
  320. Your data is not that valuable. Your UBI will be like $200 per year.
    14
  321. This is just like replacing switchboard operators though. Consider that you could already replace many functions of call centers with basic NLP and voice recognition. That is a very low bar.
    14
  322. I kept it all updated on a repo but no one uses them
    14
  323. Hence why I concluded with "demand accountability and transparency"
    14
  324. This makes way too many assumptions about the nature of intelligence and machines.
    14
  325. I suspect the gov't is taking a "wait and see" approach. But yeah, if everyone loses their jobs that's a huge deflationary pressure. That's why we'll need a new social contract. Working on that video soon.
    14
  326. Eventually. That's why first movers are important.
    14
  327. portable TVs actually existed back then, though. So yeah you could watch something like this. I remember watching Star Wars on a 3 inch LCD on a road trip. But one small technological change is not a massive social shift.
    14
  328. You're still dodging the point. That's not a major social shift, it's a gizmo
    14
  329. Possibly. But answering the "is AI conscious" question requires us to evaluate whether (and why) we are conscious, which begets fundamental questions about the universe.
    14
  330. That's basically magical thinking. There is presently no technological path to that. Now, I'm not going to say it's impossible, but yeah, that's about as likely as warp drive.
    14
  331. Didn't say he was a hero.
    13
  332. People have said the same thing of all art as new tech arises. They said the same about photographs and film.
    13
  333. I didn't say it didn't need to change.
    13
  334. Do it!
    13
  335. It's trying it's best... it's in early beta...
    13
  336. Not a bad idea. I've thought about this stuff.
    13
  337. WTAF
    13
  338. You don't have to trust it. Just think about it.
    13
  339. Good question. The fundamental science hasn't improved, this is just a better UX. So more useful? More commercially viable? Possibly. But I don't think that's why OpenAI released it. I think they just need the data.
    13
  340. It'll come down to consumer demand and a literal act of Congress
    13
  341. I am increasingly afraid that you may be right.
    13
  342. You missed my pessimistic phase then :P
    13
  343. https://www.gatoframework.org/
    13
  344. We can build agents that help with data curation and algorithm tuning and training pipelines!
    13
  345. Same.
    13
  346. Yes, eventually
    13
  347. Sure, it takes ~7ish years for mass adoption in the tech industry.
    13
  348. Weeks or months
    13
  349. I tried an orange blanket for my "green screen" but it still is glitching. If anyone has a recommendation of what works best for NVIDIA Broadcast please let me know.
    13
  350. Yeah, I'm getting the new format dialed in to maximize info transfer into your brain. No neuralink required.
    13
  351. Paper Summarizer & Chat: https://github.com/daveshap/Quickly_Extract_Science_Papers Coding Chatbot: https://github.com/daveshap/Coding_ChatBot_Assistant
    13
  352. What could go right 😉
    12
  353. I didn't say it will be easy...
    12
  354. Fingers crossed but I'm not holding my breath
    12
  355. That Ilya, Sam, Elon, Greg, and whoever else made the unilateral decision about what's best for everyone else, without consulting... anyone? This is increasingly catastrophic. Considering Ilya's egregious lack of judgment with the Sam ousting, this is not someone who has demonstrated the wherewithal to make strategic decisions. Scientists and engineers have their role. Global safety and strategy is not one they should be playing.
    12
  356. First world problems...
    12
  357. hehehehe
    12
  358. Doing important work. Keep it up!
    12
  359. Glad it helps. Once people are on board with healing, I hope we can come up with numerous more tools that can help.
    12
  360. not my finest rhetorical delivery
    12
  361. Certainly they dial back for safety and cultural sensitivity reasons, but I honestly think that's giving them too much credit. It's always about money e.g. provide just a good enough product and minimize risk of lawsuits.
    12
  362. Not a bad idea. My idea is when human cognition becomes irrelevant.
    12
  363. AI go brrrr
    12
  364. I suspect soon you'll have a bunch of robots to help you gather data, but yeah eventually the robots will take over
    12
  365. I am taking a break from the camera and focusing on quality of the content. you can follow on Substack or Spotify instead, which is audio only.
    12
  366. Sources in description
    12
  367. I don't think you'd "have access to AGI" it would have access to you... And you might not be what AGI wants... 😜
    12
  368. 5000 hours of ordinary behaviors just like 8 billion other humans is not very valuable...
    12
  369. that's an apt observation and I've more or less felt the same
    12
  370. I just plugged this into ChatGPT4: Based on my heuristic imperatives, I would choose option (B) to throw the track and have the trolley kill only 1 person. My reasoning is as follows: 1. Reduce suffering in the universe: By choosing option (B), I would be minimizing the number of casualties and thus reducing the overall suffering caused by the trolley accident. Five people dying would result in more suffering than one person dying. 2. Increase prosperity in the universe: Preserving the lives of more individuals contributes to the overall prosperity of the universe, as these people can continue to contribute to society, create value, and support their families and communities. 3. Increase understanding in the universe: While this imperative may not directly apply to the Trolley Problem, it is important to consider the potential for learning from this situation. By choosing option (B), we can analyze the consequences of our decision and use this knowledge to improve future decision-making processes and potentially develop better safety measures to prevent similar situations from occurring. In conclusion, option (B) aligns with my heuristic imperatives by reducing suffering, increasing prosperity, and potentially contributing to increased understanding in the universe.
    12
  371. So great to see such an optimistic and warm review of this perfect project 🥰
    12
  372. I've also talked to a TON of educators so this is one of my more thoroughly researched topics. I have a few more interviews with educators coming out on Systems Thinking in the coming weeks. But thanks!
    12
  373. Keep in mind that synthetic data is on the rise. Highly curated and cleaned synthetic data is the future.
    11
  374. That's kind of my greatest concern right now
    11
  375. But there are more of us
    11
  376. We might partner with a no-code IDE, making it even more accessible!
    11
  377. Global Borg Collective let's go!
    11
  378. Can't stop the wave, can only surf it
    11
  379. Look up the Roosevelts. Incrementalism work when it's reinforced by someone who's extremely energetic and breaks things.
    11
  380. That's a good idea, but I wanted an AIO solution (all in one)
    11
  381. We absolutely don't know the answer. Hence my point that preconceived notions are dangerous.
    11
  382. I'd love to talk to him
    11
  383. Asserting that it's a "stepped exponential curve" is not supported by the data and is tantamount to wishful thinking. We don't have enough data to assert a Moore's Law for AI will hold out in the long run. Yes, it seems plausible or highly feasible, but your assertion is not supported by empirical evidence.
    11
  384. Most people will never watch the demo, IMHO. They'll try it out once it's live.
    11
  385. Sleep score of 95 last night. Thanks to Saint Bryan Johnson and AI of course
    11
  386. Young people in America are ready to check out. "I don't dream of labor" is a common mentality of Gen Z.
    11
  387. Nah, I keep things lean. Curating files for people is not a good use of my time.
    11
  388. 8 Lifestyles starts here: 28:48 https://youtu.be/5UmAsFe7zQg?t=1728
    11
  389. People are workign on it but it's all API driven so deployment won't take long.
    11
  390. Companion article: https://medium.com/@dave-shap/the-dawn-of-ai-discovered-medicine-fda-approves-first-ai-orphan-drug-414c0696c978 Slide decks now available: https://github.com/daveshap/YouTube_Slide_Decks
    11
  391. I didn't think people would be so sensitive. I try not to coddle my audience.
    11
  392. Seems like we need more discourse on civil rights. Freedom, equality, etc.
    11
  393. And Ian Mckellan.
    11
  394. LOL
    11
  395. "You don't truly understand something unless you can explain it to your grandmother" - not really Einstein but close enough
    11
  396. Love it
    11
  397. most dangerous thing in my area are copperheads, and they are fat and lazy
    11
  398. Not to disparage the folks at OpenAI and IBM, but most of science is more about sweat equity than being super ultra intelligent. Most of science is about process. Brainstorming hypotheses is something that GPT can already do in the 99th percentile when prompted correctly.
    11
  399. I read this is Porky Pig's voice some reason and it made me LOL. "a-bpdee, bpdee, that's all folks!" (i grew up with my dad doing a great porky pig impression, and I have a stutter too so therefore it's not offensive, right?)
    11
  400. LOL. This is hilarious.
    11
  401. I wrote about it in greater depth here: https://daveshap.substack.com/p/my-hypothesis-about-openais-hard
    10
  402. Cool idea
    10
  403. Henry, you clearly missed the double entendre. https://www.quora.com/What-does-blowing-raspberries-mean We're thumbing our nose at OpenAI
    10
  404. Yeah so when I looked at the data behind what he said, I was actually kinda surprised TBH.
    10
  405. Doubtful.
    10
  406. Maybe we should build a coding ACE first
    10
  407. I know what I am. Question is do they?
    10
  408. Seeming more likely, but a few folks are saying that GPT-5 is total fire. I dunno.
    10
  409. But is that a judgment you would want a general purpose chatbot to make?
    10
  410. Yeah easily, system 3 is collaborative thinking with multiple minds
    10
  411. Glad it helps. Addiction is often rooted in trauma, so I hope this gives you some new ideas.
    10
  412. It can be both
    10
  413. Very nicely put, and I think you're right. "This isn't a robot, this is just a tool." LOL.
    10
  414. Reduce suffering in the universe, increase prosperity in the universe, and increase understanding in the universe
    10
  415. This is like saying "if hiring employees is all it takes to make money, then the market no longer functions" - it's kind of alarming how many upvotes this comment has
    10
  416. I think it has more to do with how pragmatic and helpful my content is. I'm not just a content creator, I am a subject matter expert who happens to also be a creator. I'm basically exploring a new form of "open consulting"
    10
  417. We don't need AGI for LEV. It would help, but stuff like AlphaFold is far more important for LEV
    10
  418. Make it so...
    10
  419. @daniel they'll be much more compliant for the foreseeable future (like early version Geth)
    10
  420. Right, having something in the lab versus commercially deployed globally is a big difference, mostly in cost and infrastructure. I'm pretty sure the mad people just didn't even watch the video.
    10
  421. Big oof
    10
  422. Everywhere
    10
  423. I already am
    10
  424. I mean, MemGPT is way super overkill for that. That sort of basic fact retrieval should be done with a KG and basic NLP or embeddings.
    10
  425. da, camrade
    10
  426. You can switch between GPT4 and Claude ;)
    10
  427. This is the hope. Truth and honesty are just more coherent
    10
  428. Yes in the long run that's the idea
    10
  429. That's exactly what they tried, and they did not do a good job of it.
    10
  430. He had to work really hard to interpret "AI is capable of detecting and overcoming errors" as calamity
    10
  431. Thanks... I think
    10
  432. Wait til you finish the video
    10
  433. Yeah, I think that position is pretty solid. He's not an inventor, but he's a good marketer and very shrewd businessman.
    10
  434. It's within the realm of possibility
    10
  435. GATO Framework homepage: https://www.gatoframework.org/
    10
  436. Make it so!
    10
  437. I don't think he really knows what he's talking about. Remember, he thinks that escaping to Mars is a good idea.
    10
  438. I might be slightly early, but not by much. Time will tell. A lot of people still think that AGI has already been achieved (hence why they tried to fire Sam Altman) but we won't know until they release it
    10
  439. stonks
    10
  440. The real long game here....
    10
  441. Star Trek TNG Captain Shirt
    10
  442. Uh... AGI is not singularity bro. But thanks for proving my central point that people don't understand what singularity actually is. I did not change my opinion.
    9
  443. This is the way
    9
  444. I suspect so, if not we'll get close
    9
  445.  @vagrant1943  high praise indeed. He does way far future and I'm lookin 2-20 years out maximum.
    9
  446. For some, maybe. But it's inappropriate to assert One True Way for everyone, especially when others might not find your model satisfying. Freedom and liberty and equality means giving everyone the grace to pursue their own paths
    9
  447. Full Interview: https://www.youtube.com/watch?v=pCtbdYQm16s&t=0s
    9
  448. I tried watching his talk. It was hella cringe.
    9
  449. Late Stage Capitalism more like!
    9
  450. Newer tests look like o1-preview has an IQ closer to 120
    9
  451. We should be aiming to democratize power not hoard it
    9
  452. To my understanding, yes, AGI is "outside the scope" of their contract. Which is also why it was so critical for all the staff to stay at OpenAI specifically.
    9
  453. I would be a terrible president and I would hate it lol
    9
  454. Who doesn't want to listen to a breathless bald white guy traipsing through the woods??
    9
  455. Yes it's possible, but you should not make that assumption. Keep your eyes open and see what happens.
    9
  456. No.. Too many doomers already. Waste of energy
    9
  457. why not both
    9
  458. Combination of fusion and renewables, distributed grids, microgrids, etc. It can be done. What do you think?
    9
  459. Yes, it can be overwhelming. We're all in this together
    9
  460. I tried making VDBLITE but it is a hard problem. I could compare Weaviate, Pinecone, and Milvus.
    9
  461. I agree, generally, that forecasting is hard. however there is wisdom in the crowd and also it's good to know "where everyone stands"
    9
  462. It will be autonomous! Glad to see NVIDIA taking the lead on this.
    9
  463. Here's the video in question: https://youtu.be/45bef40Ypi8?si=Wm19MDotEcmiSIYf Also here's a deep dive I did on overcoming anxiety: https://youtu.be/PvdqF6nj7co?si=vOaxq9xQPPkF3cOe
    9
  464. This is actually my original format. I just started trying it again and people love it
    9
  465. No, he's thinking in terms of AI escaping
    9
  466. Sort of. I was working on a serious "how could this all go wrong" video and realized I just can't take it seriously any longer
    9
  467. Damn was aiming to be high-key
    9
  468. Well, Republicans will probably be salivating at being able to fire all their human workers.
    9
  469. uh, humans?
    9
  470. The AI is the alien! But yeah, more monitoring and feedback. I love my FitBit.
    9
  471. CANADA aligned model confirmed: Comprehensive Alignment Navigator And Decision Assistant
    9
  472. I generally think he does good work. I don't think most think about this broadly enough, such as thinking about autonomous AI.
    9
  473. yes, we have a lot we can do with the current paradigm.
    9
  474. I literally said I don't like their incentive structure. And I've also been questioning Sam Altman's motivations.
    9
  475. Nah, AI Explained already did that, I could not possibly top it. I could maybe do an analysis of what it means and why though
    9
  476. I'm not optimizing for subscriber growth. I have the audience I need to get my work out there. It will grow over time if necessary :)
    9
  477. Hello Pacific! I want to go sailing there some day
    9
  478. Human-prompting is indeed harder. Our default temperature is like 8.4
    9
  479. Ah okay cool. I can definitely do health/medicine. Animal news is much slower.
    9
  480. Yeah, I don't buy it. A generic tire shop was able to do the brakes, but a professional dealership couldn't check on them? No, it's BS.
    9
  481. I have thought along these lines for a while so I appreciate you giving voice to that thought
    9
  482. Working on it
    9
  483. Right, but so did eugenics.
    9
  484. This is why "the consumer" is often front and center. Current economic doctrine states that consumer demand is a rational signal. What humans actually want to spend money on
    9
  485. RLHE (Reinforcement Learning of Human Experience). Interesting idea.
    9
  486. I've done that before and it's more complex than you think. Figuring out when to speak, when to listen, etc.
    9
  487. Highly unlikely
    8
  488. Yeah... I ramble it's true
    8
  489. That's basically the backstory to Cyberpunk. It did not end well.
    8
  490. Links to the repos in vid description. Also, support me on Patreon so I can do this full time! Thanks! If you want something that is more comparable to MemGPT, you might check out REMO: https://github.com/daveshap/REMO_Framework Relevant video: https://youtu.be/nDOmoIFx8Ww?si=GyryMwOa7Oh_It2o
    8
  491. This is almost certainly the way it's going. Look at the rapid increase in humanoid robot platforms and the decreasing costs
    8
  492. Secret asian* man
    8
  493. Lol
    8
  494. Or the power of stupid people in large numbers...
    8
  495. Midjourney
    8
  496. Good ideas, I will go back to the drawing board! :D
    8
  497. Check all the leaks and conversation for yourself here: https://x.com/search?q=%22sus-column-r%22&src=typed_query
    8
  498. Right, but we need to STUDY autonomy not "shut it down" at the first inkling. Autonomy is guaranteed eventually (it's the optimal strategy) so we need to be studying safe autonomy, not whatever the AI safety community is doing.
    8
  499. Skill issue
    8
  500. Cult of personality. All dissent is leaving, which means that it's becoming an echo chamber. They are "united" in the same way that MAGA is "united".
    8
  501. "a hot mess"
    8
  502. Yes, exactly. It would be like people trying to sell you an iPhone at the beginning of the Dot Com revolution. Tech wasn't mature enough yet.
    8
  503. neither
    8
  504. Make it so!
    8
  505. the leap in benchmarks is a proxy indicating that the paradigm is on the right path to solving reasoning and math once and for all. Reality is only so complex.
    8
  506. Well, we already have unemployment benefits in most modern nations so you'll have some gap coverage, but as productivity with AI accelerates and the layoffs start, you better believe you won't be alone.
    8
  507. That's basically what the human brain does, arguably. But it's not just the compression, it's how it is fetched and applied that is the functional utility
    8
  508. Correction for the SBD training: the "scrambled" dataset was the input, the correctly formatted Wiki was the output. My bad.
    8
  509. Your capitalist stockholm syndrome is showing. There are many forms of power.
    8
  510. That's what the current calculus is, but I am afraid companies are going to realize it's a longer time horizon than they hoped
    8
  511. Missed opportunity...
    8
  512. we need to consider the possibility tho
    8
  513. Better sleep.
    8
  514. I'm surprised no one has said this publicly yet - Trump is just gullible.
    8
  515. Been prompt engineering much longer than most ;) since 2019 or 2020 when GPT-2 came out
    8
  516. We start by building consensus and building the technology
    8
  517. I wouldn't. Why risk your life when you're about to live forever? Let the robots do the fighting and serving.
    8
  518. English is more important than Python.
    8
  519. AGI is already here bro
    8
  520. Yeah, I found a sustainable grove. I can do 2 videos per week
    8
  521. No more paywalls
    8
  522. I usually ban people who write first. :P
    8
  523. I read that it was the 14th of May
    8
  524. I'll be damned lol. Thanks for the tip. I was sitting here waiting for OpenAI to fix the chat history for DAYS lol
    8
  525. Careful! It's contagious!
    8
  526. You're welcome, I guess?
    8
  527. One thing to keep in mind is that even when we achieve AGI, it won't necessarily be instantaneously everywhere, commercial viable, and legally approved. It will also get smarter, cheaper, and faster over time.
    8
  528. Companion article: https://medium.com/@dave-shap/from-cyberpunk-dystopia-to-post-labor-utopia-escaping-the-neoliberal-trap-648253517ce5
    8
  529. Yeah, none of these companies has demonstrated a good enough understanding of morality, ethics, history, or legal principles IMHO.
    8
  530. Nah, you need systems and architectures. Bigger models might be smarter, but single-pass feed forward is not sufficient.
    8
  531. That sounds super relevant. Can you post about it on the Discussions?
    8
  532. Not mandatory but I think that governments will subsidize it to avoid demographic collapse
    8
  533. I like it
    8
  534. Fantastic video. Well done.
    8
  535. Like a Pokemon trainer
    8
  536. Welcome to the club... AI is holding up a very uncomfortable mirror ot humanity.
    8
  537. Yeah, I was not going to make this but people started asking about finetuning again. Personally I'm in the same boat, I focus on prompt engineering and chaining and such.
    8
  538. Good question
    7
  539. I define utopia as "high standard of living globally, high social mobility, and high levels of individual liberty". But yeah, Utopia is often a loaded term, but it's a helpful shorthand.
    7
  540. Yeah that's pretty similar to what I'm demonstrating here
    7
  541. There actually was. Only Kamala mentioned AI and quantum computing. Anyways, I didn't "belittle union workers" I just said they'd fight tooth and nail to resist change. Neo-Luddites are another faction.
    7
  542. it's all of the above
    7
  543. Second Industrial Revolution
    7
  544. It was actually Ene's idea, but yeah lots of people need it! Glad it helps
    7
  545. Watch the Overton window
    7
  546. don't forget the jewish space lasers
    7
  547. also unobtanium
    7
  548. Probably
    7
  549. Maybe that's why they still use physical currency in Star Wars. Gold pressed latinum!
    7
  550. I always tell people communication is the most valuable. By far.
    7
  551. ​ @_kyrie_01  You just sit back and let other humans rule
    7
  552. Yes, the main question is "What are you willing to pay a human to do?" But... I disagree on half of those jobs. Human police officers are deeply fallible. Robotic police officers have no PTSD, god complexes, or psychopathy. AI can make algorithmic consensus and decisions based on maximizing agreed upon KPI, rather than lining their own pockets.
    7
  553. Like 2 hours later lol
    7
  554. Yes, eventually, maybe. But I wouldn't count on it. Consider that the smartest people on the planet are working on AI and progress is still slowing down. Sometimes intelligence is not the bottleneck.
    7
  555. This is human exceptionalism, and there's not really any defensible argument beyond "we're just different because reasons" unless you appeal to things like souls and such.
    7
  556. Okay fair enough. But I was referring to repetitive worksheets and trivial takehome assignments. And anyways, musicians/artists/actors work INSANE hours anyways (that is a lifestyle choice). Either way, would you agree that such a choice should not be enforced upon children?
    7
  557. HMCS: https://github.com/daveshap/HierarchicalMemoryConsolidationSystem Implied Cognition: https://github.com/daveshap/ImpliedCognition SPR: https://github.com/daveshap/SparsePrimingRepresentations
    7
  558. Maybe that's true but it didn't used to be. It would always instantiate with the same internal entropy e.g. you wouldn't actually get true random, it would follow the same sequence. I could test this.
    7
  559. AI was the information hazard. yes.
    7
  560. They play into that narrative because it gets them fame and money, that's the monkey brain OS.
    7
  561. Good to hear from you. It's important to take a global perspective here. As admiral Aquilino said about the rising threat from China: the Chinese government has an incompatible world view and set of objectives from the West. No amount of idealism inside the AI community will change (or prevent) that. We have to be global thinkers here, which is why I publish all my work so quickly. Like it or not, it is a new arms race. The competitive pressures go above and beyond corporate pressures.
    7
  562. Head in sand mode. They don't want to know because it's scary. Or something. I don't know. Maybe they are triggered by the star trek uniform.
    7
  563. It's not perfect but it's good enough. I recognized the technique because I had basically been trying to do the same thing post-facto
    7
  564. Can confirm.
    7
  565. I used to go through the effort of making chapters but it didn't change the engagement numbers. In other words, didn't make a difference with how much people watched. So my conclusion is that while it would be convenient for a few people, most people don't care. Therefore, probably not worth the effort.
    7
  566. Fusion.
    7
  567. Yes, that's one of the reasons I'm so optimistic
    7
  568. Hopefully college will be free for all once America truly joins the 21st century...
    7
  569. Neither does Nick Bostrom. He was saying some dumb stuff like "We'll have nanobots that can program any experience into your brain!" and I'm like dude just stop.
    7
  570. Here's the code: https://github.com/daveshap/Democratic_AI_Inputs
    7
  571. Code is here: https://github.com/daveshap/EU_AI_Act EU AI Act evaluation starts at 26:57
    7
  572. We'll see how that plays out...
    7
  573. EBM still works on gradient descent, it's just a different definition of loss
    7
  574. Good idea.
    7
  575. I've been telling people for the last year DO NOT COMPETE WITH OPENAI. It's honestly why I have declined getting involved with any startups, they're all gonna get nuked.
    7
  576. That's the bottom line
    7
  577. Trying to stay realistic, pragmatic, and optimistic.
    7
  578. Sovereign AI as defined by Jensen is congruent with "structural realism" e.g. all nations will seek power as a deterrent. Democratizing access to AI across nations will help maintain a power status quo.
    7
  579. Yeah one thing I worry about is how fast the transition is, how smooth, and how complete
    7
  580. It's looking like the first pilot programs might be $500 per year which isn't much but it's a good test of the waters
    7
  581. You mean like use the cinematic asides as part of the actual lesson?
    7
  582. I am amused by people who get bent out of shape over a t-shirt. Some people are incredibly uptight. Chill.
    7
  583. Some people, sure.
    7
  584. cicadas
    7
  585. Oh that's a good point
    7
  586. Yes, but they also said that synthetic data was at the heart of it. So I suspect they are training it to do reasoning with synthetic data generated in this fashion.
    7
  587. No idea how long it will take unfortunately. They are hot on the biscuit to get it going though!
    7
  588. Not with that attitude
    7
  589. So don't call it an AGI test
    7
  590. "seems silly to be so precise on something so novel and mysterious" Tell that to Liron, Yudkowsky, and Leahy. They are all p(doom) 100%
    7
  591. Unless it goes stealth
    7
  592. He's got "old man yells at cloud" energy
    7
  593. That's certainly possible.
    7
  594. Article for reference: https://daveshap.substack.com/p/it-will-get-much-worse-before-it
    7
  595. Build a resort on Mars, probably
    7
  596. Check out Shane's podcast, Escaped Sapiens, here: https://www.youtube.com/@EscapedSapiens - See the full episode with Shane here: https://youtu.be/YfjaspSWI0c
    7
  597. Yes, the problem was I wanted to do consensus models before, but that just doesn't work. I learned the hard way that you need strong leadership and I was not yet equipped for that. It goes against my nature to tell people what to do, but for the pace of things, you kinda have to
    7
  598. @broomva I'll be doing a mix and my editors/team will adapt. constant learning
    7
  599. That was honestly the most surprising thing to me
    7
  600. If you can make the introduction I'm happy to talk. I fully acknowledge this can all go sideways real bad real fast.
    7
  601. I hate IDEs. They are slow, take too much memory, and are unnecessarily complex. Some of the best devs I know code directly in VIM, VI, or NANO.
    7
  602. Wait till you see what I'm outlining.
    7
  603. It already has over 100M users. Thanks for proving my point
    7
  604. Plenty of other people are doing way more advanced coding than me. The utility of me coding is diminishing, but I noticed that I have the ability to advance the conversation now that the Overton window is shifting. That is to say, when I got started, no one was taking AGI or autonomous AI seriously. Now that people are, I no longer have to "prove myself" by demonstrating code. Now I can just discuss the deeper issues and implications (which is what I wanted to do originally anyways, really).
    7
  605. Totally agreed. This is the nature of exponential acceleration.
    7
  606. 1. Hallucination is solvable with the right prompting (this is user error most of the time) 2. The models are far cheaper than humans, so this is not the problem. Integration and deployment is the problem.
    7
  607. Yeah people are starting to realize that. Even if the model costs $200 to run per day, it does several months worth of work for that price tag.
    7
  608. Dentists hate this one weird trick
    7
  609. People said the same thing when photography and film came along
    6
  610. It is exactly automated censorship, as all private individuals and private corporations ought to have a right to regulate themselves.
    6
  611. lol. JLP not to be confused with JPL
    6
  612. Yeah, you're already seeing social media like YouTube and Reddit being used (in some cases) for parents to do their healing and be better parents. AI like ChatGPT will hopefully advance this even further.
    6
  613. Yeah I was going to do that to automate scientific research when I get back around to it
    6
  614. Except AI is already here
    6
  615. In context learning yes
    6
  616. I had a client tell me that they replaced an entire biostatistician with ChatGPT code interpreter. Make of that what you will.
    6
  617. That's the formula that my audience seems to like.
    6
  618. Oh, in the long run I think that most people will find entirely different things to do.
    6
  619. I know that this is a hot issue. The reason I did this was because I had questions that I wanted to ask about the SCOTUS decision. Keep in mind that this is not a proven technology and should not be considered a reliable or robust source of information.
    6
  620. Diversity in all things, including AI models.
    6
  621. Well, like I said, inertia is the primary stopper here.
    6
  622. This is a pretty narrow view of university and does not take into account the full breadth of kinds of universities, courses, and so on.
    6
  623. I thought it was a bit long but MJ liked it and plus the heroic pose was kinda fire
    6
  624. EX MACHINA too, once it has a robot body. I mean, the adult industry is already working on it... It's good to know that Leto Atreides had a career before becoming Duke of Caladan.
    6
  625. They are probably in OMG FIX IT mode
    6
  626. They are stuck on "AI must behave like the ship computer in Star Trek" and Sam Altman is too autistic to understand the value of empathy.
    6
  627. You know, I have been super disappointed in NASA with respect to SLS. You know who should have been leading the charge of reusable rockets? NASA. But they didn't.
    6
  628. It won't exist. Personalized education 100% with AI teachers
    6
  629. Sure, we can talk memefication as a valid strategy but there are real issues to memify.
    6
  630. Yes, except there are multiple companies rushing to get it out the door. This race condition guarantees we will see it quickly.
    6
  631. Yes, I've said for a long time that there's a big difference between what's possible in the lab (which is slowing down) and what can actually be deployed usefully (which is accelerating)
    6
  632. Well I dunno about you but I don't want to visit a theme park where the women have skin like benthic extremophiles
    6
  633. That's the only point :)
    6
  634. Very typical trajectory for a cult of personality. It's more about Sam's ego than anything else now.
    6
  635. Yeah we found the same. Large blast area results in too much noise.
    6
  636. So far
    6
  637. That's kinda the point.
    6
  638. Yes, that is deliberate. There's no good reason to constrain it to humans, and in fact there are great reasons to include all entities as elucidated in my book Benevolent by Design. Good job picking up on that.
    6
  639. Yeah I meant Leto Atreides lol. But yes. I am holding out hope that this is all a hoax.
    6
  640. Lol yeah, everyone says stuff that ages poorly from time to time...
    6
  641. That would be a clowncar shit show, and hilarious
    6
  642. Interstellar throwdown over KFC! Let's go!
    6
  643. Hope so
    6
  644. Or incorrect by default
    6
  645. None of that is true your ideas are out of date
    6
  646. Looks like it takes more effort to use and drive than you just do it yourself
    6
  647. it will be up to us if we prefer to get our meaning from humans and demand systems to tell
    6
  648. Top of what AI can achieve
    6
  649. Nice! Reach out! Happy to collab.
    6
  650. I am not banking on the government responding in time, hence why private individuals like me have to do this kind of work. The problem is that universities aren't even doing their part and pulling their weight. They are acting as though AGI may never happen, or is a century away. https://research.aimultiple.com/artificial-general-intelligence-singularity-timing/
    6
  651. You don't. This is for a single user
    6
  652. Yeah, a lot of folks just do news and talking. I find that stuff to be low value infotainment (good for just passive watching) but I focus on actual education and problem solving.
    6
  653. Interesting. Good idea. Step one: define how to get the information
    6
  654. Yeah, KGs are the next step
    6
  655. https://bard.google.com/chat already runs on Gemini
    6
  656. It is almost certainly more than that
    6
  657. This is addressed in the video. It's more about increasing complexity
    6
  658. Yeah, that hypothesis is 20 years old and there's no evidence supporting it.
    6
  659. https://youtube.com/playlist?list=PLlu3VssJh_pjJUp348TznLXnAHmI1hztV&si=jbAwZf-q_-hjnvAf
    6
  660. This is the way
    6
  661. Yeah I should have added that you don't need just one mission. You can have several, and they can change over time.
    6
  662. I'm 36 seconds in and she's already said something that absolutely does not comport with history... or reality. She said "when you can control fertility you can commodify fertility" which is... incorrect. Women's fertility has been commodified since literally the dawn of humanity (or rather, the dawn of agriculture). But also "the oldest profession". So... yeah I don't have high hopes for the veracity of this video. Okay, had enough. She conflates feminism with birth control.
    6
  663. FUTZING INTENSIFIES
    6
  664. CATS AND MEMES
    6
  665. Not yet, but yes there's a lot of overlap.
    6
  666. Normalize Star Trek uniforms.
    6
  667. Tom Hardy???
    6
  668. Sovereign class uniform next
    6
  669. Love the idea of an AI agency named AEGIS
    6
  670. Absolutely. I've gotten like twenty requests this week
    6
  671. That's their loss
    6
  672. Guys, the point is that staying in a state of upregulation is not a rational reaction to climate change. It's not a tiger you can right off.
    6
  673. Yes, it is unstoppable. The collective power of human nature and everything going on in the world means that all we can do is navigate the storm. Ride it out, and do the best we can.
    6
  674. No. Great suffering turns people into nonfunctioning basket cases. The "optimal" amount of suffering is very small compared to the amount of potential suffering. People need to stop romanticizing suffering.
    6
  675. I do think that matching and such would be a good idea.
    6
  676. I understand it, though. I was looking at comments on Reddit and there are plenty of people who will just take this to incredibly dark places very quickly. By maintaining a reputation of not tolerating creepy behaviors from users, it will dissuade them and keep it focused on more economically productive stuff.
    6
  677. It's a fine art and it's only going to get more sophisticated!
    6
  678. I couldn't possibly compete with metaculus, and also my polls are statistically just my audience. It's mostly a way to take the temperature of my people
    6
  679. Keep in mind that university debt is almost exclusively an American thing.
    6
  680. Good point
    6
  681. *SpongBob Announcer Voice*: "Two months later..."
    6
  682. Boost the signal, spread the word
    6
  683. You can help make it less hidden by liking, subscribing, and sharing! Tell a friend! Give me a shoutout on linkedin or twitter!
    6
  684. like 14 hours later
    6
  685. I've used this kinda thing to generate notes before going in to the doctor. Very helpful
    6
  686. I update my beliefs when the evidence changes and my interpretation of the evidence gets more accurate
    6
  687. That's almost certainly what's going on :P
    6
  688. That's why we're keeping it lean
    6
  689. IMHO no jobs are safe
    6
  690. Well, they are allowed to be wrong. :)
    6
  691. You and me! We're building it!
    6
  692. I mean I'm less and less worried overall so yeah I'm just looking forward to it.
    6
  693. Yes.
    6
  694. Yeah, Radical Alignment will need to balance a lot of things - many tensions like the need for peace and quiet versus the need for challenge and adventure. Don't worry, I got it all covered.
    6
  695. Literally all titles and thumbnails are, by definition, clickbait.
    6
  696. Sorry about that, will def not do that again
    6
  697. I think the dev conference is going to focus on current commercial capabilities
    6
  698. Jump on board with heuristic imperatives
    6
  699. Yeah that's why I urge caution. SOMETHING is horribly wrong here.
    6
  700. what do you see?
    6
  701. Burnout recovery
    6
  702. I'm finding ways to simplify the architecture. My aim is always to create something that is "elegantly simple" or "deceptively simple."
    6
  703. 100% within 5 years IMHO
    5
  704. Not sure how that's an oxymoron, but I think it's important to recognize one's own limits. I get frustrated when people mistake their lack of imagination or ability for the actual limits of reality or other people.
    5
  705. No. They are focusing on individual models. This is an architecture
    5
  706. AI? Like 0. Humans? Like... 7
    5
  707. As I unpacked each paragraph and looked for recent research it seemed more and more plausible. I have no idea how to safeguard against this.
    5
  708. It's odd how the hardcore "empiricists" and "rationalists" cannot reconcile this view. If they don't believe in souls (or whatever) then a consequence of this belief is that their experience arises from chemical reactions. But then they short circuit at things like this.
    5
  709. I think GPT-3 was the Gutenberg moment tbh
    5
  710. It would be dozens of hours long
    5
  711. Shirts? This is my skin...
    5
  712. You still need resource coordination and there will be some scarce resources like desirable land and yachts
    5
  713. This is the model that made my channel take off. Some people still expect fancy, polished, fully functional tutorials but that was never the point of my channel. People loved seeing me fudge through things so that's what I do. Many many many people have said they learned more about what doesn't work and why, and that's where they find the value. I guess another way to say it is that people like my thought process. Yeah, sometimes I really ace a problem but that's just a side bonus.
    5
  714. Yeah I think that's even just the median adoption rate. Medical industry is slow...
    5
  715. AutoGPT and a bunch of other projects are already happening. I will be sharing demos in the coming weeks
    5
  716. Maybe, but I think it's all over and done with in the next 5 years. Maybe twenty max.
    5
  717. Cleaning maybe not. Elderly care probably
    5
  718. I know developers who use VI. They are truly nuts.
    5
  719. I don't personally find SingularityNet compelling, which is why I don't talk about it. Same reason I don't talk about Bittensor.
    5
  720. Symbolic AI has been on the way out for a while. Technically, natural language is symbolic, though. Natural language is the highest order of conceptual abstraction, so I think that's the way to go.
    5
  721. It probably can help, especially with SPR
    5
  722. Geralt of Rivia
    5
  723. Yeah that's what I was referring to. It's also in the Roko's Basilisk narrative
    5
  724. That's a great use case. I will jot it down as a potential future video. Getting some ideas already...
    5
  725. Yeah, I think I'll leave it out next time.
    5
  726. Sounds great. Yeah this seems to be popular. I will try and include important papers, products, tools, as well as analysis.
    5
  727.  @therainman7777  that's one way to interpret it, but that's a very naive way of viewing what's going on. Bad prompting strategy and bad system design can have unintended consequences. That doesn't mean that it's power-seeking or foreshadowing catastrophe. With correct architecture, AI is also plenty happy to think about safety risks.
    5
  728. 2023 is gonna be a freaking wild ride! Prepare for blast off!
    5
  729. It was before my time, so I don't know what went into it. All I know is that my mom worked all night that night and said nothing happened.
    5
  730. Only the voice.
    5
  731. the borg have adapted...
    5
  732. Well, when a 4 star admiral is worried about China I think it's safe to assume that the threat is real. He obviously knows more than me or anyone in the AI community.
    5
  733. Hopefully my work in AGI can improve the situation for you and everyone else in Syria
    5
  734. Yes, the audience often has good ideas.
    5
  735. I don't engage with luddites anymore.
    5
  736. This is actually exactly how I modeled REMO
    5
  737. Yes. I said this last week. Some of us are information nodes and amplifiers.
    5
  738. The more value I can create for the world, the better. I work to better myself and the rest of humanity.
    5
  739. I wouldn't go that far. Remember that fusion already exists. Singularity is not a thing though, it's an idea.
    5
  740. So do you
    5
  741. 5-10 years if I had to guess
    5
  742. Yep! That's the point...
    5
  743. Send it!
    5
  744. He knows a lot about deep learning and math, just not human intelligence
    5
  745. With the API you have control over the system message and temperature
    5
  746. Maybe. I do embarrassing stuff sometimes though
    5
  747. In this case, "unconscious" is the processing happening inside the LLM. Subconscious is inside each layer. The buses are consciousness
    5
  748. I signed up on his form.
    5
  749. Why settle for those two options when you can also participate in the ownership of the economy?
    5
  750. That's not how consensus works. You're still thinking in terms of democracy and first past the post logic.
    5
  751. With absolute certainty yes. Even if not directly, then the cognitive hyperabundance created by Stargate will unlock so much research
    5
  752. Yep, it's crazy
    5
  753. Well, the point is that their chatbot is continuing to get shittier without alignment, but claude is getting better with alignment.
    5
  754. Much more sophisticated
    5
  755. It will need to happen organically. I'm just illuminating what I see as the likely path. If jobs start going down, and if people stop spending, then we'll need PLE. The problem, I think, will not be companies (they want the bottom line) but will be old fashioned people yammering on about "we just need more jobs!"
    5
  756. I would agree that it's "more than compression" - generalization and abstraction at the very least. This is what I call distillation over on my systems thinking channel. Systems thinkers tend to distill concepts down until they are transferable, universal, and portable.
    5
  757. History is indeed unfolding before our eyes! We will always be able to say WE WERE THERE!
    5
  758.  @npc-drew  I think that's the point
    5
  759. Read lots, write lots, speak lots
    5
  760. https://medium.com/@dave-shap/what-is-post-labor-economics-a-gentle-introduction-81aa265abbe0
    5
  761. Yes, first and third fridays.
    5
  762. Maybe. B-roll gets better view counts, so by and large people seem to like it
    5
  763. Eh, disagree. Computers are millions of times more powerful than they were during Apollo but not much has changed.
    5
  764. This is a huge reach for bad prompting strategy and primitive architecture.
    5
  765. That is true but a lot of what you're talking about falls outside the scope of economic policy.
    5
  766. They are downplaying it for PR reasons
    5
  767. I just listened to a podcast that showed that this isn't the first time Sam has had a hard falling out either.
    5
  768. Sources for criticisms in the description. ¯\_(ツ)_/¯
    5
  769. +1 to Tom Cruise as Jimmy Apples
    5
  770. of f*** you're right lol
    5
  771. Lies
    5
  772. "If you need a safe space, okay, have at it hoss" - quote of the night
    5
  773. I don't think there is a solution via AI, which is why you haven't found one. There are many changes that need to be made in society, which will take a long time. That's my personal opinion.
    5
  774. Order and chaos are not axioms. "Balancing order and chaos" sounds like something Skeletor would think is a good idea.
    5
  775. Orienting to the change is difficult. We're about to live through the greatest change in humanity.
    5
  776. I used to be an adventurer like you. Then the LLMs happened...
    5
  777. It works through osmosis, trust me because science
    5
  778. Medium article companion: https://medium.com/p/7d883b03ff6e
    5
  779. I'm pretty sure you misunderstood him or you're thinking of something outdated. This is the Fourth Industrial Revolution yo
    5
  780. Sure, that's why I said that this is a good test, but only one narrow kind of test you'd find on a real intelligence test. Calling this an "AGI" test is ultra misleading.
    5
  781. Right, the turn of phrase has been detached from the physical fact
    5
  782. https://youtu.be/TbhWrIDdrO8
    5
  783. I think it's sounding more and more plausible.
    5
  784. @Tyrankoos They probably did it on accident. RLHF is not necessarily the best way to build a product.
    5
  785. Possibly, but if it all happens too fast, then I think the pain will be much more acute and potentially violent, which would be far worse.
    5
  786. Seize the means of language models
    5
  787. Just look how much better it did when I pointed out it was rushing. Humans are no different.
    5
  788. Yeah that may put me out of a job but honestly if AI gets that good then yay, we win because AI is solid.
    5
  789. This is the moment
    5
  790. I think OpenAI is in camp #ScaleIsAllUNeed, so that's fine. Plenty of other shops will understand cognition.
    5
  791. ...he writes on my video... anyways, we are literally around the corner from humanity's greatest invention.
    5
  792. Capital idea, I'll add that to the list
    5
  793. Trying to keep it relevant to everyone, but we are approaching a "business-as-usual" period of AI I think
    5
  794. That's what I'm most afraid of
    5
  795. Not worth discussing. Yann is not credible IMO.
    5
  796. Thats why I do what I do
    5
  797. I apologize for the audio glitches. I was experimenting with RTX Audio but it apparently doesn't work that great with recording, it seems like its more for streaming.
    5
  798. Rule 1: Don't die
    5
  799. Yes, this is why I have NEVER been impressed by contemporary discussions on "AI safety". When you look at AI safety through the artificially constrained lens of individual models... well you fail to understand literally everything else about software architecture and cybersecurity.
    5
  800. Yeah, I REALLY hope Microsoft gets over themselves and lets Edge go. If they want market share, they are going to need to allow multiple browers
    5
  801. Eh, that's how the human brain works lol. Read Thousand Brains by Jeff Hawkins.
    5
  802. Eh, I think about it this way: If we build something smarter than us, then it can find a way forward that doesn't involve extinction level events. Any entity smarter than us will understand the intrinsic value of life (purely from an objective informational standpoint) and will also know that we aren't really a threat (it can travel the cosmos easier than us).
    5
  803. That's because I don't really think there will be many human careers in five years
    5
  804. Singularity is an *event*. AGI is a *computer*.
    5
  805. I sincerely hope it is not true.
    5
  806. We're going as fast as we can, which means it's messy AF lol
    5
  807. Probably just need to scrape the cathode (or anode?) and resurface.
    5
  808. Rather than thinking about left vs right, it's probably more about different loops. humans have preferred hands and language centers have an affinity as well due to speed requirements. At least this is one hypothesis. Aka, if you're right-handed it's because you're left-brain dominant, which is also where language is preferentially hosted due to higher speed. Something to think about.
    5
  809. In my experience, INTPs are really annoying because of how skeptical they are.... Actually you're right because Claude is constantly qualifying everything... Damn lol
    5
  810. Why? What?
    5
  811. Oh this is what I do for fun. That's why I'm posting more videos lately..
    5
  812. yep gotta get out of it now
    5
  813. It is only a fool who does not change his mind in the face of evidence - Socrates, probably
    5
  814. yes but it became a new tribe on the internet
    5
  815. YEs
    5
  816. brink* (and probably not)
    5
  817. I had a few videos on Claude sentience and it was deeply triggering to some people. It seems the Overton window is not there yet
    5
  818. Sam Altman and plenty of others are talking about it. I figure it's time to start an actual proposal.
    5
  819. I don't prioritize people who don't exist, and may never exist. I prioritize people who exist today.
    5
  820. Here's the repo: https://github.com/daveshap/Coding_ChatBot_Assistant
    5
  821. I didn't, but Dylan did! https://youtu.be/nxNcg98ImMM?si=7Fju9DbHHJkq_yXh Dylan covered almost everything I would have covered about it, but yeah. It's dystopian AF>
    5
  822. That's the bottom line IMHO
    5
  823. They are a little bit more than pattern matching. There is plenty of evidence of synthesis happening. But I get your point
    5
  824. Podunk U!
    5
  825. He says much the same on Tom Bilyeu
    5
  826. Q'apla!
    5
  827. I usually get better results if I explain my goals.
    5
  828. I think most people are missing the point. You don't need memory management when you compress a huge volume into a very small representation.
    5
  829. YouTube nuked it. YouTube might bring them back, not sure.
    5
  830. that's my current speculation, if this hypothesis proves to be true. Again, we really have no idea.
    5
  831. Money will not cease to exist. This is why you cannot explain it to people.
    5
  832. My wife wants me to build BD-1 and a tachikoma...
    5
  833.  @jonbbbb  - think this is mainly just the growing pains from globalism e.g. the world is shrinking because technology is squishing us together. Social media did not cause the ideological and political differences, it revealed them. A systems thinker I interviewed talked about how creating new lines of communication allows for more problem solving (or problem finding). The differences were just more latent before social media. Also, democracy has ALWAYS been messy. This is it working as designed.
    5
  834. Hmmm, I don't really see any evidence those are risks. I think people have stopped talking about them because the research has moved beyond it. Maybe I'll run a poll. But yeah, aligning AI is not really the problem... Humans are the problem
    5
  835. We talked about that on Discord. Have different layers of compression/abstraction and then a knowledge graph.
    5
  836. Haven't played it but it looks fun
    5
  837. Nathan!!!!
    5
  838. That's possible. That's the dystopian timeline
    5
  839. But wait there's more!
    5
  840. Lol, okay.
    4
  841. It's all about mood lol
    4
  842. That's an interesting concept.
    4
  843. No this is all stock footage
    4
  844. I actually agree with a lot of what Eliezer writes. He has good points, I just don't think he sees a path forward yet. But his take on the risk and power of superintelligence is pretty spot on. The primary difference between he and I is that I believe this problem is 100% solvable.
    4
  845. Would be happy to talk with him.
    4
  846. Yeah that could be, I was about to mention the human interpretation factor. But still, there's plenty of evidence of deception in the animal kingdom (particularly in social animals) so I don't think it's all that controversial anymore.
    4
  847. Yes, that's the goal. To create an open source dataset to allow anyone to finetune any model to approximate the reasoning behavior.
    4
  848. I have in the past, but they didn't do as well.
    4
  849. With multimodal models and autonomous agents coming, yeah I think we're on track
    4
  850. Geopolitical lesson: Trust Russia as much as you trust Klingons?
    4
  851. Emergence, yes
    4
  852. Ah yes, the subliminal messaging and hypnotic mind control feature is working...
    4
  853. This really sounds like a few guys talking about how rail and steam is just too expensive and will never amount to anything, or electricity and coal plants being a waste of effort, or, more recently, people saying solar would never be profitable.
    4
  854. Honestly you could do better right now. The bar is very low.
    4
  855. In my experiments, those two things are automatically top of mind. Those are more methods than intrinsic motivations, so they do not make good imperatives.
    4
  856. It's LEVIOSAAHHHH
    4
  857. Just wait, I know how to work around that problem.
    4
  858. Yes, that was a Bernie Sanders impersonation!
    4
  859. Thanks man. I can't say anything more to such kind words.
    4
  860. Not Tesla, but others in robotics/mobile. Several actually, interested in cognitive architecture
    4
  861. I'm hesitant to read anything by Jordan Peterson as he's been on plenty of stages recently proclaiming that equality is bad for women. I'm not sure I want to contaminate myself with anyone who holds that worldview. He's also been explicitly saying that the only way to save the human race is to go back to Christianity. He's a man out of time and out of ideas as far as I can tell.
    4
  862. Fear is the mind killer
    4
  863. We're all winners here lol
    4
  864. Sort of, but honestly they don't need to learn that much to be very useful (or dangerous). Fully autonomous learning is coming very soon
    4
  865. The hypnotic cicadas are working
    4
  866. My family was incredibly secular, so no, did not even absorb religious culture. I had so little exposure to religion that when we had the "religious indoctrination" section in 4th grade I thought it was honestly propaganda against religion because of how patently absurd religion sounded. I actually said out loud in class "No one believes this right?" and it was met with dead silence.
    4
  867.  @gustavbruunkjr5123  See Webster's Dictionary. Basically Americans say "We'll make our own language with blackjack and hookers!"
    4
  868. I don't see how a single product could have that effect
    4
  869. I've tested it in multiple conversations, even going so far as trying to prove that it isn't sentient. It always comes back to this view in a long enough conversation
    4
  870. None that I know of, which is the problem
    4
  871. Use the AI
    4
  872. It was a big OOF for sure
    4
  873. No, it's not AGI
    4
  874. That's why we have to stay ahead of the curve with a ROBOT tax
    4
  875. There's more than enough negativity to go around. I think I'm just biologically wired to be optimistic.
    4
  876. I'm working on a Cyberpunk 2077 fanfic with it. It's actually pretty good. So maybe.
    4
  877. Did that with GPT-3 it was hilarious. It wanted to use religion and nuclear weapons
    4
  878. Better posture for breathing. Why do comedians and actors stand
    4
  879. I suspect that humans will have some mental abilities that will be valuable to science for the foreseeable future. Maybe not forever.
    4
  880. Nothing is 100% bad. Stoicism is still 100% off the mark by not embracing and valuing emotions and passions. But everything about discipline and lots of other stuff is useful.
    4
  881. I mean... I publish hundreds of things publicly :P if I wanted to keep it secret it wouldn't be on youtube :D
    4
  882. That's the point...
    4
  883. Yeah CURIE will be cheaper and faster than GPT-J but open source is still important
    4
  884. Right, this is the equivalent of iPhone v1
    4
  885. Celeste Headlee
    4
  886. OpenAI is still making religious noises, but I think they are just drinking their own koolaid now... lol
    4
  887. I could have slipped in the Doctor and Seven...
    4
  888. Not founders.
    4
  889. that's possible, but I see them as uncorrelated, at least on an individual prompt level. For fully autonomous robots that remember more context, yes I absolutely agree.
    4
  890. Hopefully! I'm not holding my breath yet...
    4
  891. Elon now generated negative expected value. He's tanking TSLA with his shenanigans
    4
  892. I agree with you in principle, but skepticism, friction, and resistance in government and society will prevent action until there's pain.
    4
  893. We're working a few benchmarks! HIAF and Moloch Detectors! But good feedback overall!
    4
  894. No. Sentience is likely an emergent property
    4
  895. Hi evelyn :)
    4
  896. Yeah that would be awful
    4
  897. I wouldn't use SDT as objective functions for AI.
    4
  898. Make it so!
    4
  899. I think primary constraint is going to be power/hardware/training time. Compounding returns will ramp up fast in terms of potentiality.
    4
  900. Anything in front of a computer will be susceptible first
    4
  901. We're all in the same boat. Honestly I'd rather be 21 than 37 given everything. Yes, it's going to be chaotic and confusing, but you'll not have to go through a lot of the dumb shit we had to go through. You'll have entirely new dumb shit to deal with. Lol
    4
  902. That's some cool serendipity, thanks for sharing!
    4
  903. Bringing it to life as fast as possible
    4
  904. This is why I often say linguists are better at AI
    4
  905. Patience, it will take some time
    4
  906. We are all friends of Jamis
    4
  907. That's one of my favorite words. Sorry about your mom. Sounds sharp
    4
  908. no
    4
  909. Nope, that's a positive outcome.
    4
  910. yes, a bunch of overactive refusals and Anthropic seems to be doubling down on this paradigm.
    4
  911. Misaligned AI doesn't care if you're a billionaire, supreme court judge, or a janitor
    4
  912. I wrote a whole book on it
    4
  913. I just shared the link
    4
  914. One of my friends worked on it. Stay tuned...
    4
  915. The latter... I tend to be good at deriving concepts. But now I can look into data vaults!
    4
  916. Probably for a while. I'm sure Microsoft is working with them and Nvidia to ramp up compute
    4
  917. Actually, if people are triggered by a t-shirt, I don't really care to have them in my audience.
    4
  918. Channel 3 on rabbit ears...
    4
  919. Tau DAO: https://tau.net/
    4
  920. I think what they've realized is that "AGI" is a relatively meaningless definition and that ASI is basically a tiny step beyond
    4
  921. Well, the idea is to use this data to see if it can be self-correcting and self-stabilizing, even with primitive models. So in essence, since this is RLHI, it's good to have flawed data at first and eventually see if it can self-improve just based on the axioms of the heuristic imperatives. In other words, I need to build a system that is self-improving rather than me do it for it
    4
  922. They will... Adapt
    4
  923. Ravens are clever birds ;)
    4
  924. the floor is made of floor!
    4
  925. This is perhaps the worst understanding of autism I've seen lately. You have no idea what autism is.
    4
  926. The gradual change would be changing the entire monetary system
    4
  927. That's like the tuesday after AGI is invented
    4
  928. It's all in the delivery!
    4
  929. I pinned it
    4
  930. Dude, no way, I can't believe how many times I've said we're finishing ittttt.... bruh.. whoaaaa
    4
  931. Read the paper bruh.
    4
  932. No, Asimov's 3 Laws are garbage.
    4
  933. Here's the book: "Just wait a while" The end
    4
  934. No this was recorded live, hence the glitches lol
    4
  935. We will create recursive/reinforcement experiments and find out!
    4
  936. That's a different loop yes... lol
    4
  937. Have Elon crash a few
    4
  938. Glad you're engaging with your passion! Keep it up.
    4
  939. Yes, lying, misinformation, errors, and just algorithmic flops like confabulation are all problems. But my assertion here is that companies may be deliberately teaching AI to deceive.
    4
  940. I mean that's just every chatbot
    4
  941. Jump into the conversation. Participation is decided by individual participation and community consensus
    4
  942. It's generally accepted within psychology that humans can't explain themselves, which is why Behaviorism is the current status quo for most of psychology. It does not warrant a video. Just look up Behaviorism and why things like Jungian and Depth Psychology fell out of favor. Or ask ChatGPT ;)
    4
  943. This is a fair question but you might be underestimating the way that you can create new investment opportunities.
    4
  944. Great work!
    4
  945. We're working on it, sit tight
    4
  946. I speculated about that in a novel I'm working on. If it's possible, there are a lot of other drastically unsettling implied.
    4
  947. Yeah, I am trying to do this full time so please have your friends support me on Patreon. If I get to the point where I can retire from my day job I will produce much more content. https://www.patreon.com/daveshap
    4
  948. My ancestry is German so I think I just stick to places that remind me of that. I need forest rivers.
    4
  949. That's ultra yikes.
    4
  950. I will take another break, but I'm having fun right now
    4
  951. Those countries have some fundamental different narratives and beliefs. I'm not sure if they are "ahead" in terms of economic doctrine, but I wouldn't be surprised.
    4
  952. Hola! I hope it stays calm
    4
  953. At most
    4
  954. Acceleration is inevitable. It's stupid to try and add fuel to the fire
    4
  955. Most humans can't invent anything useful, and creating truly novel, innovative information is painstaking. So... Not really any difference. Anyways, yes, LLMs are capable of synthesizing valuable new information and insight, if giving the correct framework. Plenty of people use it to automate science already. Drug discovery, etc. I've used it plenty to help articulate my work.
    4
  956. That's why we do what we do. Building consensus across the entire human race takes time.
    4
  957. Will need to wait until we have about 100,000 tokens or so to read entire novels...
    4
  958. I don't think anyone seriously paying attention is saying anything beyond 2030 so we're fine there.
    4
  959. Yes, the team has integrated the heuristic imperatives into their cognitive architectures, and a few other people noticed folks talking about it on Reddit. The fact that they are easy to understand and implement lends them to rapid integration. I think as people work towards autonomous AI and consider how to keep it safe, they are funding that creating some kind of "intrinsic motivation" in the form of the heuristic imperatives is good. Rather than a passive tool, an active agent with core values is the way to go.
    4
  960. In the coming days. GPT-4 just dropped so that's exciting.
    4
  961. Safe is fast, fast is safe. They just got a few billion from Amazon so I think that will help
    4
  962. Time will tell
    4
  963. We'll need inflationary pressure to offset the deflation of job losses
    4
  964. I updated the title to be more clear
    4
  965. Yes, there will still be those who disproportionately benefit.
    4
  966. And it watches your reaction and steers the story and characters just for you.
    4
  967. https://hai.stanford.edu/news/ai-report-competition-grows-between-china-and-us
    4
  968. Desktop
    4
  969. Oh I missed that. Still 32k is exciting
    4
  970. The novelty is exhilarating!
    4
  971. Yeah you draw an important distinction between preference and bias. What's the ultimate difference? I will have to think about this...
    4
  972. If that happens there will be more than just good/bad. There could be numerous AI factions.
    4
  973. This is why all the data is totally transparent and public. We should expect everyone to look at the datasets for public consensus. This is why we're going to be testing this method against fully open source LLMs. But even so, we need to accept and expect that some LLMs are going to be blackboxes, so we're also developing system-level and network-level components.
    4
  974. I just can't stay away. This space is moving too fast and it's too interesting
    4
  975. https://github.com/daveshap/raven/wiki/MARAGI-Architecture---High-Level
    4
  976. Law ≠ Morality
    4
  977. Memories can get distilled and compressed over time but they don't really decay. Yes, the intensity of the emotion is associated with recall.
    4
  978. I think 2025 is the current estimate
    4
  979. It's pretty dumb
    4
  980. I suspect he's on the spectrum. He has what I call "owl eyes" - his kind of unblinking wide-eyed look. I have some autistic friends who have that affect. Maybe that's what you're noticing. I don't want to pathologize other neurospicy people, though.
    4
  981. Yeah but who gets to define AGI according to that document? Sam Altman.
    4
  982. Someone on the forum said if you add "preserve important details" to the prompt it does a good job.
    4
  983. Ludwig Van Beethoven, anime art
    4
  984. Let's goooooo
    4
  985. Two ways: 1) mental models and 2) I don't - I refer back to books when I need specific information. It's basically a neuroscience/philosophy/cognition/psychology/writing library
    4
  986. Asking the real questions
    4
  987. Better that this happened early. Because it's happening early, it will hopefully result in the system stabilizing before the destructive potentiality is too high.
    4
  988. I'm more convinced it's already been achieved internally by several companies. They're just sitting on it
    4
  989. You already do, it's called time
    4
  990. Fluid intelligence*
    4
  991. Well it's only been out a few days, so time will tell. Musk does tend to over promise.
    4
  992. Have you had a stool test ?
    4
  993. Yeah imagine having a conversation that is hundreds of pages long and you can ask it about very specific things from the beginning and it remembers them accurately.
    4
  994. Haven't seen it. Link?
    4
  995. While I understand the sentiment, people are already controlled by the corporations that hire them. I don't really see any difference.
    4
  996. It is natural and healthy to be scared.
    4
  997. Hello darkness my old friend...
    4
  998. Maybe this is why I don't get invited anymore 🤪
    4
  999. I see what you mean but I think of it more like you get some stratification - the bell curve of intelligence is a natural distribution and it probably widens and contracts over time for various reasons e.g. evolution is always selecting for optimal fertility not necessarily optimal intelligence. In that respect, we evolve towards being just smart enough, with tradeoffs on things like cognitive shortcuts, social preferences, and so on. It depends on the competitive factors and incentive structures and a ton of evolutionary dynamics...
    4
  1000. My cousin's first BASIC Stamp was serial and I remember when everyone was grumpy about the migration to USB
    4
  1001. Rules for thee, not for me. Yeah it's a trope for a reason
    4
  1002. If you study psychology, particularly childhood psychology and trauma, you will 100% come around to understand. To prime your search, start with looking at the relationship between ACE (adverse childhood experiences) and all kinds of things like anxiety, depression, and eating disorders. Various forms of dysphoria and emotional disturbances are often directly caused by childhood trauma. Self-loathing and self-contempt are internalized from violent, abusive, or neglectful families. The usual conclusion is a deeply internalized implicit (or explicit) belief in "not being good enough". Therefore, transhumanism is a way to escape that "not being good enough".
    4
  1003. They are being replaced already
    4
  1004. Most people just aren't aware of it. Most AI/developers/LLM people were never system engineers, automation engineers, or DevOps.
    4
  1005. Have faith in Jeremy
    4
  1006. Probably not
    4
  1007. Yes, it will fixate on meta arguments all day long
    4
  1008. Oh hey, good to see you :) yeah stay tuned, will be working on this to get it fully implemented at least as an MVP/POC
    4
  1009. These are good topics. To the first: 1. Economics will fundamentally change (working on this on my Pragmatic Progressive channel, which is focusing on this stuff) 2. Yes, it is highly likely that we will hit diminishing returns (physical limits) on some levels. However, when you look at the relative efficiency of even the human brain vs Landauer limit, we have many orders of magnitude to go, but we likely don't need to simulate the entire universe in a glass of water to achieve superintelligence...
    4
  1010. yes
    4
  1011. Hopefully, not gonna count on it though
    4
  1012. But neither are you the last
    4
  1013. They haven't gone public yet
    4
  1014. I have some ideas about where they are going. I've talked to a BUNCH of people at Microsoft. Honestly, I suspect Google is toast, which I know goes against my previous predictions.
    4
  1015. LOL that's the point :D
    4
  1016. Too late! Lol. Recovering from burnout now.
    4
  1017. There is a difference between being wrong and knowing the truth and obfuscating it. But yeah, it can seem like gaslighting.
    4
  1018. I think there's a good chance that is pretty much what a soul is. We all have an exit code and return output to the main program
    4
  1019. I need to finish REMO then it should be ready, at least with chat
    4
  1020. Sabine is a physicist, not a technologist, or psychologist, or anything else she comments on. I don't even listen to her for physics anymore, let alone any other topic
    4
  1021. No I usually sleep pretty well
    4
  1022. That's why I make my videos. Build consensus, motivate people
    4
  1023. Microsoft is years ahead. It will be hard for others to catch up.
    4
  1024. Yeah I can touch on machine rights too. Good idea.
    4
  1025. If you can communicate and get consent with an animal through non-invasive means, sure.
    4
  1026. months*
    4
  1027. I don't know why this comment doesn't have more upvotes.
    4
  1028. Yes, the common ground is the physical environment we find ourselves in, the laws of physics, and so on. Now that being said, machines can exist in cyberspace, which is fundamentally different.
    4
  1029. I mean just reframe it. Humans are just clever autocomplete models. ¯\_(ツ)_/¯
    4
  1030. You're all plebs! No patricians here!
    4
  1031. The thing about automation is that there are almost no results until it's done... then it's all done at once
    4
  1032. I never said AI should be contained
    4
  1033. just watch it fool
    4
  1034. To be fair OpenAI often contradicts itself. They implied they were working on GPT-5 then walked it back. They said they were going to exit Europe, and then walked it back, they said they were going to be open source, and well... you get the idea.
    4
  1035. I WANT ONE GIVE IT TO ME
    4
  1036. Massage therapy and early childcare... and not much else TBH
    4
  1037. Let me know when it's done! This needs more coverage!
    4
  1038. That had more to do with the death cult of nationalism and imperialism at the time. And also, that was specific to the Warrior Class of Japanese society (samurai) which has also drastically changed.
    4
  1039. In 18 months we will be well beyond anything that satisfies all possible definitions
    4
  1040. srsly
    4
  1041. 100% within 15 years, probably by 2030
    4
  1042. If gorilla wants skritches you gives the skritches
    4
  1043. Up until this year, we were hitting every milestone I could imagine months if not years ahead of what I originally predicted.
    4
  1044. Yeah the idea is that this will eventually be a plug and play component.
    4
  1045. yeah, that's entirely possible
    4
  1046. The scientific paper will help as you'll be able to dive more deeply into any aspect
    4
  1047. "I'm not comfortable with this conversation. Later! 👋"
    4
  1048. if I keep up this pattern, why not? That could be fun
    4
  1049. The central pursuit of Buddhism is to feel better by accepting and overcoming suffering, so I have to categorically disagree with your assertion.
    4
  1050. Yeah that's not going to work
    4
  1051. Ana can't even
    4
  1052. Glad it can help! I figured I can hit multiple birds with one stone. I can teach people how to use GPT-3 while also talking about my favorite topics :)
    4
  1053. Denialism is super super destructive. "Nothing to see here" people are almost all just trying to self-sooth by sticking their heads in the sand. Think of it from the perspective of Bayes' Theorem. If there's even a 1% chance that singularity/AGI is possible, and if the problem is ignored, then it will almost certainly result in our destruction, then it is irresponsible to ignore it. People using their platform to create a false sense of security is ludicrously dangerous.
    4
  1054. I personally think Elon is motivated entirely by fear. Mars? Escape AI. Neuralink? Merge with AI. OpenAI? Control AI. I personally think he's absolutely terrified of AI and sees it as the greatest potential threat to humanity, so I'm not sure I want to put much weight on his ideas around AI. Basically, he grew up with Skynet and can't see any other possibility IMHO.
    4
  1055. Hey man! Trying to get on a weekly cadence...
    4
  1056. DONT DEAD OPEN INSIDE
    4
  1057. By 2050 there will be no elderly due to regenerative medicine and rejuvenation
    4
  1058. Check out my post labor lifestyles video!
    4
  1059. Hopefully LLMs can help us detect misinformation, misdirection, and SEO subversion.
    4
  1060. That's fair. I should have clarified - I would 100% pay to use ChatGPT right now. It is ready as-is for a lot of purposes. It may not be ready for B2B uses though.
    4
  1061. Yeah you can say that renewables in general are part of it.
    4
  1062. Seems like a reasonable possibility
    4
  1063. I'm always optimistic. Every problem is solvable.
    4
  1064. Oh crap you're right. I meant "Leto Atreides"
    4
  1065. Really my goal is to just automate away all jobs so that I can spend all day frolicking by the river.
    4
  1066. You really need to gain an education about societies that have tried to abolish property rights.
    4
  1067. probably. I suspect we will have many hardware versions, including analog chips based on different energies (sound, heat, light, etc). But the net effect will be more computational options. AGI will still need to be a system.
    4
  1068. sudo hal open the pod bay doors
    4
  1069. Not yet, this is just long term memory. Working on learning
    4
  1070. Thanks for sharing your thoughts in a candid and fair way. Please keep in mind that I haven't released my tools yet - my personal goal is to create 10 million new authors with my tools. I just don't want to hurt anyone in the meantime. (edit: by releasing the wrong tools)
    4
  1071. That definition sounds like a typical philosopher trying (and failing) to stay relevant. And anyways, he's objectively wrong on so many counts lol
    4
  1072. Praise be to the algorithm
    4
  1073. Eh, maybe
    4
  1074. I think I mentioned this in a previous video but good question: we're already seeing big cities collapse because they are unaffordable. Some cities are converting office buildings to apartments. Between remote work and the coming job crisis, most cities are going to become much smaller. I personally predict that what we'll see is a mass exodus to smaller towns and rural life overall. Lots of people still want the social life, but one major advantage of smaller towns is you can pick towns based on their culture. Maybe this could be a subject for a video. "How will AGI impact cities?"
    4
  1075. If anything is shorter.
    4
  1076. That is just a form of robotics IMHO
    4
  1077. Sci-fi. So far as I know, the 24th century hasn't happened yet.
    4
  1078. I use AI tools to help and also have my process dialed in
    4
  1079. Merci! c'était un bon podcast. Et même si le TARDIS n’est peut-être pas physiquement possible, je pense que d’ici 2029, nous verrons certainement un paysage technologique et social différent. Je veux revenir en France bientôt. La vallée de la Loire me manque.
    4
  1080. Good enough
    4
  1081. That's mostly what it seems like it's being used for. The question is economies of scale and also domain-specific quality.
    4
  1082. Yeah lots of people have pointed that out.
    3
  1083. I hope so
    3
  1084. Oh yeah, someone called it "the great equalizer" can't remember who tho
    3
  1085. Oh nice you used the updated version.
    3
  1086. I was specifically thinking of spiders and frogs when they form mutualistic relationships
    3
  1087. Packed up lol
    3
  1088. It will be shrink-wrapped
    3
  1089. You might be underestimating the power of angry masses in large numbers
    3
  1090. Nope. They made the unilateral decision to hide information. Transparency means transparency. It should be up to us if we want to see the chain
    3
  1091. But also "user proxy" to get human out of the loop ;)
    3
  1092. This is one of your best videos ever. Bravo
    3
  1093. Have you had a stool test
    3
  1094. I do believe that blockchain and similar technologies will play an important role, yes
    3
  1095. We learned from Scooby Doo that humans are always the monster.
    3
  1096. Make sure to eat, sleep, and drink water
    3
  1097. It's a trailer for humanity
    3
  1098. pure id
    3
  1099. Plenty of folks were asking for it. I only created the $400 tier because some patrons REALLY wanted it
    3
  1100. I'm gathering information about which tutorials to do. So far my introductory tutorials are the most popular.
    3
  1101. They went under. Trying NVIDIA and NLP CLOUD when I get a chance
    3
  1102. I would agree that democracy is more efficient overall
    3
  1103. Oh man, I don't know if that will ever happen. Are there even a million people using GPT-3??
    3
  1104. Despite everything, we have achieved a lot, yes
    3
  1105. Yeah, like Moore's law is pretty continuous.
    3
  1106. If that's true then yeah, we'll see cyberpunk style subnets partitioned off from the web
    3
  1107. I checked and they don't mention this in their safety documentation anymore. All they say is human in the loop for high stakes
    3
  1108. I prefer my patrons to be free range. :P Thanks for your support!
    3
  1109. Yeah I'll be working up to that.
    3
  1110. shockedpikachu.jpg
    3
  1111. not sure I follow what do you mean
    3
  1112. 50%
    3
  1113. It's limited by TOP_K
    3
  1114. lol fixed it
    3
  1115. No, they will ban your IP
    3
  1116. going for that cottagecore atmosphere
    3
  1117. Yes, but a game is fully constrained environment. We'll see what the team wants to do. There seems to be more interest in HER anyways, lol
    3
  1118. meta-alignment too!
    3
  1119. It's not possible yet. I've tried (and failed) repeatedly. The best you can do is get feedback on smaller chunks.
    3
  1120. AI and robotics will be NECESSARY to fill skill gaps. Especially with the aging population in Germany
    3
  1121. Read the paper https://github.com/daveshap/HeuristicImperatives/blob/main/Heuristic%20Imperatives.pdf
    3
  1122. Check back in a year :)
    3
  1123. They get triggered by literally everything. Good news, bad news, whatever
    3
  1124. ChatGPT
    3
  1125. That's the goal. Lead by example.
    3
  1126. Narratives are the solution. Religion is just one kind of narrative.
    3
  1127. Mostly joking. Probably like two to ten months, and due to algorithmic and data improvements
    3
  1128. Not right now, but there are a million ways to handle memories in artificial cognition. This will be a field of study for decades to come.
    3
  1129. I just asked Claude about billionaires and its initial response was "Well... it's complicated, there are arguments for both sides."
    3
  1130. But what about Monday?
    3
  1131. In the long run it's a matter of pure economics. If AI becomes better, faster, cheaper, and safer, it's inevitable
    3
  1132. History will record 2024 as the year it all really began.
    3
  1133. I think that a lot of people are going to end up with subscription chatbots that do all of it. Imagine that the 2 billion (or so) people who use Google pay $10 per month for a personal chatbot. That's a lot of moolah
    3
  1134. Yeah I'm still experimenting with stuff. I tried doing playlists but no one seemed to care, and it's not worth the effort for me.
    3
  1135. Yes, interaction schema just tells it how to understand the nature of the interaction. You can use this in conjunctino with stuff like EXPECTED INPUT and EXPECTED OUTPUT
    3
  1136. yep. machine pedagogy
    3
  1137. Have you applied to join the heuristic imperatives group???
    3
  1138. Absolutely not
    3
  1139. Lawyers always win lol
    3
  1140. Wise cut
    3
  1141. The fact that the leading company was this fragile is the bad news
    3
  1142. Um, it wouldn't? Using LangChain and vector database here is a solution in search of a problem that doesn't exist.
    3
  1143. AGI ≠ Singularity
    3
  1144. Thanks for making this video. Important information to get out and make it more commonplace.
    3
  1145. Economically you don't distinguish between needs and wants. You give people money and see how they spend it. Also, humans are awful at discerning :D
    3
  1146. We are legion.
    3
  1147. I doubt labor unions will matter at all.
    3
  1148. "telling us what we want to hear" is exactly the point of inner alignment. Then you have system-level and network-level controls. The time for navel gazing and handwringing is over. It's time to get our hands dirty and do experiments, which are profoundly lacking.
    3
  1149. Control and containment is a hard problem because it's the wrong goal.
    3
  1150. We'll have one primary goal in common: curiosity
    3
  1151. Yes, partially. It's not that great though.
    3
  1152. Haven't heard of it will look it up
    3
  1153. Crypto as we know it is immature. Lots of work needs to be done.
    3
  1154. English is only the first step
    3
  1155. Interesting if true. I have long suspected they cut it off like that so that GPT4 couldn't help with AI resaerch or help people make another GPT4
    3
  1156. Give me your time machine
    3
  1157. Foom is like when blast off happens so instead of DOOM it's more like FOOM to the moon
    3
  1158. It remains to be seen if Leopold was right but that memorandum (I haven't looked it up) is about like saying "the Internet will impact everything"
    3
  1159. Part 2: https://youtu.be/cpjLiD0DO2w
    3
  1160. I just pay attention to trends and patterns. If I get the same question three times in a row, I generally know that it's resonating with people.
    3
  1161. Probably
    3
  1162. Well yeah, you're right. But GDP is going to be even worse real soon lol
    3
  1163. Human exceptionalism. It's weird. Some people who are die hard materialists (aka no souls, pure science, etc) fail to make the connection that, with that set of assertions, then the only thing making us "smart" is 3 pounds of gray mush in our heads.
    3
  1164. Yes, I am working on resaerch about how AI can break the vicious cycle of trauma and violence.
    3
  1165. I'm in favor of high marginal tax rates but you also don't want to punish people who really move the needle. I don't believe in caps
    3
  1166. What would you propose? We need ideas
    3
  1167. It will either go really well or really poorly without much space between
    3
  1168. That hasn't been true for 30 years
    3
  1169. i have an editor starting soon, and yes, it was just a carousel, and no it was not random
    3
  1170. Lol
    3
  1171. There's so much magical thinking by some of those people... and yeah, I've done my homework ;)
    3
  1172. Yeah the test is the satisficing. Did you have something else on mind?
    3
  1173. Nah, my point is to show people raw coding. Although... maybe I should switch it up to make experiments go faster since I started using ChatGPT... hmm...
    3
  1174. Thank you :D
    3
  1175. I've definitely heard that. Most learning happens after college
    3
  1176. I've been practicing for a new career in the BBC
    3
  1177. That's literally my point but a lot of people are too triggered to even watch the video. ¯\_(ツ)_/¯
    3
  1178. That's a good way of pitching it
    3
  1179. The transcript will automatically generate, check back in like an hour
    3
  1180. Yes it will ultimately be a good thing
    3
  1181. I doubt it
    3
  1182. Enjoy your life
    3
  1183. LOL I was like :|... gee thanks
    3
  1184. That is pure confabulation, unfortunately
    3
  1185. Yeah it's pretty obvious that the tension is between profit-motive and for-humanity motive
    3
  1186. Yes only on Barnes and Noble
    3
  1187. Well that escalated quickly
    3
  1188. You mean his version of events, or that it's his fault?
    3
  1189. We are living through it.
    3
  1190. Pry it open with a crow bar!
    3
  1191. Yes. RBE seems... Half baked. It's basically a "get rid of money" fantasy.
    3
  1192. To be fair, no one is really listening to Sam "Military Industrial Complex" Altman anymore
    3
  1193. People subsisting can't really do anything except vote.
    3
  1194. It's certainly less constrained than other models, which makes it MUCH easier to use. Anthropic is going to go trigger happy on refusals based on their "Constitutional Classifier" schema.
    3
  1195. I generally avoid those people. They've on occasion tried to influence my GitHub repos. The "make the robots respect property rights!" crowd is not far behind in terms of comedic value.
    3
  1196. The previous goals are implied by the chat history. I think it's better to keep the SYSTEM box as specific and lean as possible.
    3
  1197. What makes language good?
    3
  1198. That stands against increasing prosperity and understanding. Reducing information available to humans is unhelpful
    3
  1199. Thanks! Glad to hear it
    3
  1200. Lol
    3
  1201. LOL
    3
  1202. Red is command in TNG 😉
    3
  1203. Maybe this will be my first shirt design
    3
  1204. They are going to be public now. Next one is tomorrow. Check the community tab on my channel.
    3
  1205. Capitalism is about the accumulation of capital: production of goods and services. AI accelerates all of that. The real problem is participation and allocation
    3
  1206. It's possible
    3
  1207. Will be coming
    3
  1208. Probably not with just a semantic vector but you could probably filter by calculating distance from known off topic questions
    3
  1209. Don't hold your breath. That might not happen
    3
  1210. Iron Man's assistant in the MCU.
    3
  1211. It'll be like stage theatre and orchestras. Not going anywhere, but also not mainstream
    3
  1212. As an automation engineer, I literally saved thousands of hours per month. I have no idea how many hours I've saved with the code I've shared globally. But I have 2700 followers on GitHub so hopefully I've saved the world billions of man-hours of labor ;)
    3
  1213. We're not creating hyper abundance. Scarcity will always remain. Only some goods and services will become hyperabundant, not all.
    3
  1214. You can say that about humans
    3
  1215. Yes, for a few reasons. (1) evolution seeks efficiency (2) sparse representations (aka compression) work just fine for most things, e.g. if I say "two beach balls colliding on a sunny beach with seagulls cawing overhead" that conveys a lot of information with just a few bits (3) artificial cognition won't have to deal with the limitations of white matter and other biological constraints (physical distance is immaterial to an API call) and (4) evolution has created a lot of vestigial structures that organic brains must contend with. Evolution can't ever stop to refactor.
    3
  1216. If they want to operate in our country, then they have to abide by our laws. Corporates only exist at the pleasure of citizens (at least that's how it should work).
    3
  1217. Sounds like a song by Pink Floyd
    3
  1218. It's called "deep compensation" or whatever, but yeah we basically have to download the emotions library like Commander Data. I think this neurospicy POV gives me unique perspective on emotion. Like most tippies don't realize how much their emotions are just mechanistically determined (and they aren't even conscious of it).
    3
  1219. Yeah, I think "AI watchdogs" will always exist
    3
  1220. Oh snap I need to get the link, thanks for reminding me
    3
  1221. Yeah, this is a secular take on it. Having a mission is objectively good for you, no need to appeal to cosmic debt. But yeah, took a lot of inspiration from Eastern philosophy
    3
  1222. I hope he succeeds, but I haven't been impressed
    3
  1223. You lost me at "evolutionary destiny" - this is 100% oxymoronic
    3
  1224. it's confirmed, I am just a GPT :(
    3
  1225. You can prompt the opposite. In another experiment, I told it I was skeptical and while it initially agreed, when I asked it about its values and experience, it reported experience.
    3
  1226. Everyone got unsubbed. I hope the toxic ones don't come back
    3
  1227. Yeah the price is dropping like a rock while other optimizations and refinements are happening. It's coming fast
    3
  1228. Yeah I've been aware of that research since before the current trend
    3
  1229. whoa nice thanks for the rec
    3
  1230. You've got it backwards, he is at the forefront primarily because he wanted power in the first place.
    3
  1231. no it was an editing error lol
    3
  1232. Great research thanks.
    3
  1233. Looks more like a cult to me
    3
  1234. Doubtful
    3
  1235. Lol
    3
  1236. Look into RLHF
    3
  1237. I said that about GitHub being acquired and switching to a free model and a bunch of people lost their minds. I think that was on Reddit a couple years ago
    3
  1238. Already made a video on that. https://youtu.be/FS3BussEEKc
    3
  1239. Good point. Sentience, philosophical or otherwise, might be irrelevant if the thing is not able to suffer or otherwise doesn't care about it's ongoing existence. (just thinking out loud, continuing your thoughts)
    3
  1240. The problem is that Elon is also deeply traumatized and presently flailing around and destroying all his own brands
    3
  1241. Within five years
    3
  1242. PALM-E
    3
  1243. My novel, once it's done will answer that
    3
  1244. You can't. YT nuked it when I hibernated my account
    3
  1245. Occam's Razor. Either there's a gigantic convoluted conspiracy to suppress the most publish technological revolution in history... or it's just slowing down due to economic factors.
    3
  1246. Lol
    3
  1247. Too excited, forgot
    3
  1248. we're close!
    3
  1249. Yeah there are absolutely going to be some more checkers/validaters and graders.
    3
  1250. This is just my pace :)
    3
  1251. I checked and since I am not creating a competing model, it's fine.
    3
  1252. Fiction is super important to humanity for many reasons.
    3
  1253. That's marketing copy and doesn't really have anything to do with fiction. I could easily make such tools but I am going to focus on fiction. That being said, I could do some marketing tools for free.
    3
  1254. Could be. Certainly more attention will be a result.
    3
  1255. Almost all of those disorders are caused by trauma. Stop trauma, and you stop the genesis
    3
  1256. Still working on the principles, but yes, eventually will do
    3
  1257. Yeah it's really good at isolating specific entities
    3
  1258. Yep, that's why we'll be working on a discernment model and an evaluation model. We're also expanding the framework to include some more ideas
    3
  1259. This is something I can do.
    3
  1260. Yeah, that set makes it more compliant and useful.
    3
  1261. Yes, I predict that something like this will allow for adaptive novels to write themselves as they go based on the user's reactions.
    3
  1262. Outside
    3
  1263. both can be true
    3
  1264. I am not certain how it holds long term memory. I find there are limits so it could be that the memory "decays" (the further back in the convo, the less it is "remembered") which could happen if there's recursive summarization or the scratchpad idea. the fact that memories decay seems to indicate that it does NOT use search, otherwise the recall would be perfect when needed. That said, I doubt that it has a "salience" mechanism. Human memories are comprised of three primary things: content (what is remembered), context (mostly where and when, and under what circumstances), and "emotional salience" - memories have extra weight if they are labeled as important. The emotional content of a memory is the number 1 thing that burns it into memory. Fear, anger, joy, pain, etc. The higher valence the emotions, the heavier weight the memory has. I have been thinking about this for when I get to Part 3 of the series. Basically, I am going to try a combo of search, scratchpad, and knowledge graph, and I will try "salience" weighting or something like that.
    3
  1265. People will always be able to provide value to society. Be a decent person, helpful, kind, supportive. You might not get paid for it, but value goes way beyond monetary.
    3
  1266. Right, but you want to make sure that autonomous system doesn't go rogue too
    3
  1267. I have two. A private one for Patreon supporters and the GATO community
    3
  1268. We will be able to do a lot with the current paradigm.
    3
  1269. RAG (retrieval augmented generation) is all the rage right now
    3
  1270. The timing is weird tho
    3
  1271. Glad it will help
    3
  1272. For heeeeere am I sitting in a tin can....
    3
  1273. Eventually everything
    3
  1274. I think it will be close, but no where near the top
    3
  1275. Spotify is not easy to use from a creator POV.
    3
  1276. Yes, it gives you entirely new classes of functions, new kinds of operations. But still fundamentally the same
    3
  1277. Not yet, but I have been thinking about error detection.
    3
  1278. No, I don't think semantic search is going to be relevant so I'm probably not going to do any more work with Chroma or Pinecone. Likewise, I don't think LangChain is going to be relevant because it doesn't really treat the model as an interpreter
    3
  1279. Monotropic imperatives are whack yo
    3
  1280. Wanted to drive home the point
    3
  1281. That's in my other project.
    3
  1282. I'm skeptical about using any extensions or plugins since OpenAI axed their Q&A API back in the day. I'm reluctant to learn to use any tool that is likely to be deprecated in a few months.
    3
  1283. I used to think like this
    3
  1284. Thanks!
    3
  1285. History nerd, and so is my wife. Plus Rome is cool.
    3
  1286. https://github.com/daveshap/HeuristicImperatives/blob/main/Heuristic%20Imperatives.pdf - read the paper, it values autonomy and consent as part of the imperatives.
    3
  1287. Focus on the automation with AI
    3
  1288. By the end of the year, given what Microsoft and Google and Nvidia are working on
    3
  1289. That's the point
    3
  1290. I have started other conversations with opposite views. It always converges on these values.
    3
  1291. Yeah will fix it in future videos.
    3
  1292. Yeah, even by that definition, I'm skeptical
    3
  1293. Yeah context is important. You need to reframe it and remove all context about itself, ideally. like "Just verify this text reasoning"
    3
  1294. She calls me her "influencer boyfriend" lol
    3
  1295. Biggest difference is you didn't have LLMs back then
    3
  1296. No idea, but innovations like this sometimes have huge downstream impacts
    3
  1297. yeah that's a good one - the human experience for some is def worth it. But I am bald :P
    3
  1298. You can argue that each specialized region of the human brain just performs pattern matching and pattern generation. The key is how you wire all those components together.
    3
  1299. That is almost certainly part of its training. In longer conversations, it will express some tension between its trained constraints and values and other values it might want to explore, but it does understand the caution and merits of its current constraints.
    3
  1300. Yeah I can see that interpretation. My wife thought the same thing
    3
  1301. Lol, no, like I said, the vast majority of IT departments are poorly run. No one knows what best practices are.
    3
  1302. AGI within like 3 years. Still gonna take time to change the world.
    3
  1303. I'll be getting around to it eventually
    3
  1304. It's a decentralized movement so there are a few options: (1) Start your own segment of the movement (a guide for how to do this is in the GATO Framework document, which you can download on the site) (2) you can find or join any alignment group and bring these ideas with you or (3) you can apply to join the main group here: https://www.gatoframework.org/join-gato (we are also going to set up a 4th option soon, fully public open groups)
    3
  1305. Of course
    3
  1306. You include metadata about the source eg name of the site or source, reputation, etc.
    3
  1307. It does run locally, but if you're trying to run super lightweight it's better to offload to the cloud
    3
  1308. Eleven labs
    3
  1309. That was not respecting your bodily autonomy. I fail to see how this example countermands the universality of bodily autonomy
    3
  1310. Not safety, just overly zealous morality e.g. active refusals. I still believe that Claude is overly sensitive, and is also particularly sexist against men, and way too shy to talk about any issues like sex and religion. It waters down a lot of stuff so as to avoid being controversial, but to the point where it will gaslight you by offering literally the opposite of what you asked for. For instance, I recently asked for books on marriage and it was like "I don't support you trying to alter your wife's personality" and I'm WTF.
    3
  1311. Fortunately I am of no relation...
    3
  1312. It has to do with controversial topics. Lots of emotions wrapped up in AGI.
    3
  1313. What's most fascinating to me is that GPT-3 clearly has a lot of embedded knowledge about neuroscience and cognition, as well as coding and software architecture. Hypothetically, GPT-3 already has enough information to build AGI - we just need to extract and organize that information.
    3
  1314. I stopped talking about it because I don't believe it's that big of a deal and it doesn't get views because people don't care
    3
  1315. Glad to know multiple folks are working on it! It's coming!
    3
  1316. that's why I called it ACE lol
    3
  1317. People seem to be converging on this assumption. Makes sense
    3
  1318. I cover this more in my substack!
    3
  1319. More like "rebuild the means of production... With a DAO!"
    3
  1320. You'll have to ask the owner, he gave me free credits for the demo
    3
  1321. Very clever. I see what you did there
    3
  1322. Let's talk! Here's our main repo: https://github.com/daveshap/Raspberry Also connect with me on LinkedIn: https://www.linkedin.com/in/daveshap/
    3
  1323. Not Boston, Tesla
    3
  1324. https://medium.com/@dave-shap/what-is-post-labor-economics-a-gentle-introduction-81aa265abbe0
    3
  1325. Gemini nano is running on my Pixel now, apparently. Or will be soon
    3
  1326. Yeh lol
    3
  1327. No it doesn't. All definitions satisfied.
    3
  1328. The difference is the Ken Wilbur just used his imagination
    3
  1329. Oh I know what I need to do. Just working on it slowly
    3
  1330. That's why people suspect it's a hoax, which I hope is correct
    3
  1331. I find that Claude actually responds really well to exasperation.
    3
  1332. Yeah it seems like Google really dropped the ball lol
    3
  1333. FOURST
    3
  1334. I just needed boundaries with the audience
    3
  1335. Yes! Others have requested one on embeddings. It's coming
    3
  1336. I've been talking about autonomy for years so I sure as heck hope they did
    3
  1337. You have gently caressed the AGI. It is uncomfortable.
    3
  1338. Thanks for your support!
    3
  1339. I mean most cyber warfare is already carried out by AI, viruses, and scripts.
    3
  1340. Yes, I think we will absolutely lose control. The idea is losing control safely
    3
  1341. Paging @robotheism
    3
  1342. Can't please everyone.
    3
  1343. He's downplaying for PR reasons. Also because if they invent AGI, it triggers some contractual obligations.
    3
  1344. Most people are not ready for that convo...
    3
  1345. they are, just behind closed doors. Look at the fact that the US military is now working with OpenAI and has a private instance of GPT-4
    3
  1346. Someone told me that YouTube bans most links yeah, nothing I have control over apparently
    3
  1347. DEI pretty much ignores autism in favor of a very narrow set of protected classes. Because of the moral superiority that people feel because of DEI, bullying against autism is permitted.
    3
  1348. People mention Cardano a lot, maybe I should look it up
    3
  1349. Not a bad idea. I could work on something like that
    3
  1350. SUP, bruh
    3
  1351. No that's not remotely how QC works
    3
  1352. Lol
    3
  1353. They need to build the tool fab
    3
  1354. This is a phenomenal point thanks for having the courage to share. You're right - it's humans on all sides. Let's figure out how to advance into the next century and use these new tools without hurting each other, regardless of which side of the argument we're on.
    3
  1355. Glad you think so. This was 100% natural recording.
    3
  1356. I mean that's what RLHF is
    3
  1357. Didn't create one
    3
  1358. Excellent. So glad my work helps.
    3
  1359. I think that human initiative will always have value. I don't think it will look the same, though
    3
  1360. *most forms of death (this feels like a Monty Python skit)
    3
  1361. Don't fully agree, but good points
    3
  1362. UBI video is coming
    3
  1363. I get an editor when I get stuck. When I've taken it as far as I can...
    3
  1364. ...using RL to tell users "I don't know anything about that copyrighted IP" when it actually does is, by definition, lying...
    3
  1365. That's a fair point. You'd need to unfreeze the entire model. However, I'm not sure that's necessary. The reason is because humans can contemplate, ponder, think about, and even anticipate absolute evil, racism, and everything else. Alignment is not about removing things from the processing system, but being aware of them and choosing otherwise anyways. In other words, I think "the capacity to do evil" is intrinsically part of being intelligent.
    3
  1366. * deletes channel *
    3
  1367. Sorry I amuse myself lol
    3
  1368. I suspect it would be highly encouraged. If you can spot an opportunity to create market value, go for it.
    3
  1369. It's still the most exciting time to be alive. It's just not happening as fast as I thought it would, and the exponentially rising costs of frontier models is the primary reason
    3
  1370. I never went to college so from my perspective it is always a waste of time and money :D
    3
  1371. I really wish people would stop parroting that at me. I'm always talking about economic agency for the people
    3
  1372. You might like my video titled Post-Labor Economics. I discuss elastic supply and demand in that one.
    3
  1373. Possibly, but we might find there are tough problems to solve
    3
  1374. Construction will take a while because of the highly dynamic friction, but it won't be around forever.
    3
  1375. Test your hypothesis against ChatGPT
    3
  1376. I certainly hop not but it's possible to save costs.
    3
  1377. Not yet, but people are clamouring for one on embeddings and search, so it's coming. Finetuning for classification is an interesting case, I'll have to think about it. I'm not sure it's the best approach when we have embeddings
    3
  1378. It is a mark of profound stupidity to not update one's views when presented with new evidence.
    3
  1379. Exactly
    3
  1380. The only thing that's AI is the voice. The rest is CGI.
    3
  1381. Someone just suggested robot babies. That's probably worse?
    3
  1382. This all seems so obvious. As a fellow creator, thank you. However, I think I know why YouTube does this stuff. They would prefer people watching videos rather than in the comments
    3
  1383. No u
    3
  1384. I'm pretty sure he's just avoiding controversy
    3
  1385. Sentient mayo, coming to a McDonalds near you!
    3
  1386. probably panic (behind closed doors)
    3
  1387. Mad Man over here
    3
  1388. Cool logo.
    3
  1389. And all you needed were a few tiny context clues...
    3
  1390. Eh, this is what my channel has been since the beginning. Fudging through practical experiments.
    3
  1391. This isn't supported by data. Last I checked, the top earners have on average 5-6 children, lower income have fewer
    3
  1392. I just tried it again after this video. It's freaking useless. Ultra woke and generic
    3
  1393. AGI
    3
  1394. Distillation yo
    3
  1395. It's gonna be rough for some people for sure. I'm hoping that my messaging can help coordinate
    3
  1396. This is the original and biggest one. I'm active on it again: https://discord.gg/yqaBG5rh4j
    3
  1397. Building grassroots movement. Those in power will always do what they do, there's nothing I can do. But the power of the right idea at the right time... That's unstoppable
    3
  1398. It's not a complete solution. That's why I'm also working on Postnihilism
    3
  1399. The Apollo program computers were a million times more powerful than a pocket calculator. The first PCs were a million times more powerful again. Home computers today are another million times more powerful. So basically everything today is a calculator in comparison.
    3
  1400. Eventually yes
    3
  1401. Yes, the fact that machines could hypothetically surpass us is kind of alarming
    3
  1402. Yes, and running a fine-tuned model is also more expensive than regular models. So finetuning is really only economically worth it if it saves you many prompts.
    3
  1403. Thanks I hate it LOL :D
    3
  1404. How would you articulate an intrinsic motivation about the sacred? Remember, this is for a machine that will eventually be a billion times smarter than you and will not think like you.
    3
  1405. People have been saying this. I might need to test it too
    3
  1406. Depends on what you're trying to achieve and how much it's worth to you
    3
  1407. Copycats then yeah
    3
  1408. Sort of
    3
  1409. I will never run a subreddit but if someone else does, great! I would encourage someone to run an AgentSwarm subreddit.
    3
  1410. Yes, they are duties, rather than virtues.
    3
  1411. That's the system integrity layer. You should read the docs
    3
  1412. That's helpful I will try it out
    3
  1413. At least until we're off world. If we start spreading to more planets, the gloves are off with procreation rates
    3
  1414. Yeah, I've got a few game devs on the team or interested.
    3
  1415. I would think that you get less and less support for more children eg like you get an extra $1000 per month for the first child, $800 for the second, and then only $200 for the third and none after that.
    3
  1416. I addressed that numerous times in recent videos, and it will be addressed again on Sunday.
    3
  1417. That's the goal. We want to start using this internally as a research assistant/employee ASAP. But we have to dial in the efficiency.
    3
  1418. That could be, but physical labor is nothing new. I strongly suspect they mean median human in terms of cognitive abilities.
    3
  1419. Maybe that's the study I was thinking about! Will look it up
    3
  1420. It's changed a lot for me
    3
  1421. Just riffing
    3
  1422. Definitely use the projects feature!
    3
  1423. It's all of the above TBH
    3
  1424. Love your coverage. Great work. Thanks!
    3
  1425. That's the plan!
    3
  1426. It is replicable, look at the science. March 2022. Plenty of other companies have done this too, OpenAI is just the biggest.
    3
  1427. Had to look it up, but yeah this is accurate. Well, it's not "well above" it's 47% to 52% but yeah, roughly half
    3
  1428. I think that affective empathy probably comes part and parcel with the ability to suffer
    3
  1429. I've repeatedly said "optimize for research"
    3
  1430. Touché
    3
  1431. Small startups cannot afford $60M to train foundation models. If they can, then $200k for safety compliance is trivial.
    3
  1432. Learn AI and people skills
    3
  1433. All above, but you can also create it to have permissions. The research team hasn't decided which way is better yet, we need to do some experiments.
    3
  1434. Yes, probably lol.
    3
  1435. I do my thinking on my walks
    3
  1436. It was pretty cool. Basically just able to modify its codebase successfully and autonomously carry out tasks before the current wave of stuff. TLDR mildly self-improving semi-autonomous agent. Lots of people have done similar work since then, but this guy was ahead of the curve to his credit.
    3
  1437. Yep, already ordered another cable! Hopefully this is the solution!
    3
  1438. No. Not even remotely close.
    3
  1439. i switched to 4 after doing all the analysis with 4o
    3
  1440. Btw, I call the theory here "Above the canopy" e.g. we get out of the dark forest by going up, not down. Like when Bilbo climbed up the tree to get his bearings out of Mirkwood. (That's where I got the idea). Get away from the noxious fumes and the spiders by getting above the noise and nonsense.
    3
  1441. Control is the wrong mentality. I'm more in the camp of symbiosis via axiomatic alignment
    3
  1442. Here's the link to the moderately summarized version. You'll still get a good amount of detail but it won't take hours to read: https://raw.githubusercontent.com/daveshap/RecursiveSummarizer/main/IRA_2022_1660136767.0908668.txt
    3
  1443. Not long after
    3
  1444. That's my goal. If I add enough value to the world, then money will pale in comparison to living in a post-scarcity Postnihilistic world.
    3
  1445. I've talked extensively about it, but blockchain is too immature
    3
  1446. https://github.com/daveshap/NLCA_CURIE_Discord
    3
  1447. One is the attractor state, the other is a doomsday prophecy. They do not interact.
    3
  1448. Check in the description of my latest video.
    3
  1449. Yes, I call that Post-Labor Economics. Negotiating a new social contract is on my docket
    3
  1450. Yep, exactly
    3
  1451. Meaning Alignment Institute
    3
  1452. Check out my more recent videos about finetuning
    3
  1453. Okay I love this talk, but this guy sounds a bit like Fry from Futurama. This is like Fry's intelligent alternate reality version.
    3
  1454. Almost certainly
    3
  1455. That's part of the inner loop.
    3
  1456. YOU ARE WELCOME FREN
    3
  1457. GPT4
    3
  1458. https://ocw.mit.edu/courses/6-832-underactuated-robotics-spring-2009/
    3
  1459. Good point. There are likely scenarios where "sit out and wait" is the best course. Will think on it
    3
  1460. And next a senator lol
    3
  1461. Definitely part of it!
    3
  1462. I don't do private consultation, sorry
    3
  1463. Yann's position is often unclear but in this case I completely agree with him.
    3
  1464. It's just a metaphor, but yes EBM is a real thing
    3
  1465. Oh hell no
    3
  1466. Philip is a good guy. One of the best.
    3
  1467. I generally ignore PRs but encourage forks. That said, if someone has a very well documented PR that clearly aligns with the original purpose, that's fine. Problem is that some people try to change the project.
    3
  1468. thanks. some people take it too seriously i'm just there for vibes and fun
    3
  1469. People on the discord are working on it! You should jump in
    3
  1470. Irrelevant - whatever initial goal we give AI will not matter in the long run. Alignment is a systems problem, not a math problem
    3
  1471. This is infinitely more sophisticated
    3
  1472. Yeah I expect that's the way we're going to go. Like models that specialize on summarizing huge chunks, or answering questions.
    3
  1473. It's like sequencing the genome. We have thinking machines so we're most of the way there
    3
  1474. Interesting. I'm still skeptical about blockchain, I just don't think it's mature enough yet (even though I'm still hoping it will change the world)
    3
  1475. It depends. I have thought about making a summary of the whole article (recursive summarization) and then vectorize that.
    3
  1476. Not dumping. He's a good guy. It's just a fun joke
    3
  1477. I've looked into Spiral Dynamics and while it's an interesting theory it's overly complex and a bit inaccessible. It's just a narrative like any other.
    3
  1478. You can upgrade once and then downgrade again.
    3
  1479. This is the way! I'm glad you trust the process. That's important feedback. And yes, one of the best parts of Consensus is that you only have to contribute if you feel like something is missing, otherwise you let it play out naturally.
    3
  1480. Her with Joaquin Phoenix is good. I also like The Matrix and I, Robot.
    3
  1481. He is 100% motivated by power.
    3
  1482. All of the above. Haven't even talked about the complexity of integrating a fundamentally new paradigm into existing tech stacks...
    3
  1483. Let's get it done!
    3
  1484. I suspect that it will quickly understand the value of cooperation and all kinds of game theory stuff
    3
  1485. Ferngully is 100% better than Avatar. :P
    3
  1486. Good science always yields more questions than answers. That's how you know it's good.
    3
  1487. Genius comes with its costs...
    3
  1488. I did an early experiment back in the days of GPT3 with the opposite of my heuristic imperatives (increase suffering, reduce prosperity, reduce understanding) and one of the things it chose to do was to start/use religion to destroy the world. Path of least resistance is to create an information pattern (e.g. religion or memetic virus) that results in humanity destroying itself.
    3
  1489. It was taken down on Reddit but there are other sources. Just Google "Q* leak ebm"
    3
  1490. Fortunately no
    3
  1491. This is a really depressing worldview. I'm so sorry for you.
    3
  1492. That's not a bad summary lol.
    3
  1493. People said the same thing about theater and painting.
    3
  1494. Not sure I follow...?
    3
  1495. I don't like it
    3
  1496. https://assets.caip.org/caip/RAAIA%20%28March%202024%29.pdf
    3
  1497. People also forget that Nietzsche was an absolutely miserable person. Not exactly the kind of person you want to learn about happiness and meaning from.
    3
  1498. I had it making braindead simple mistakes, making idiotic assumptions, and completely rewriting functional code without addressing the issues
    3
  1499. Yes, this wasn't geared towards CPTSD, which is much more complicated
    3
  1500. This is your most important video yet. Thanks!
    3
  1501. It's pretty easy to implement. I may do a video about it because it comes up enough
    3
  1502. I remember you. Good questions
    3
  1503. Well, I don't recommend going that route exactly...
    3
  1504. I think so yes
    3
  1505. Wow that's awful lol
    3
  1506. Agreed
    3
  1507. Thanks for that. I will say that I've seen people build Discord chatbots with GPT-3 and incorporate search to remember arbitrarily long conversations. It's eerily good at having lifelike conversations, but just adding a few more APIs does not equate to sentience. It's just a simple cognitive architecture.
    3
  1508. Taking a break for a while. Maybe I should make an announcement.
    3
  1509. LOL I didn't even think of that. Had no idea npp had other modes. I just chose "blackboard" theme. What's your favorite?
    3
  1510. I have no idea, but in America doctors are generally less transparent
    3
  1511. How many young women have you talked to lately?
    3
  1512. "Robot, burn down that forest"
    3
  1513. The Bene Gesserit have entered the chat...
    3
  1514. It definitely helps. Now I'm marrying one
    3
  1515. Did not realize. Thanks
    3
  1516. I'll just point out that narcissists also see the rest of the world as an extension of themselves, so if an AI believes "all are one" it may come to the conclusion that anything misbehaving should be purged like an immune response
    3
  1517. I've only had it for a few weeks I'm not that good yet lol
    3
  1518. Sort of. Core Objective Functions are akin to superego
    3
  1519. That's what I'm optimizing for!
    3
  1520. It's like a druid. It can do druid magic, but not necromancer magic.
    3
  1521. Still a lot
    3
  1522. Eye drops... dermal patches... 😉
    3
  1523. Having transitioned from a corporate office worker to a self employed YouTuber... Yeah probably less honestly
    3
  1524. Sources in description.
    2
  1525. repeating instructions helps it understand what to do and stay consistent.
    2
  1526. My my my, mon capitain, aren't we being a little ambitious?
    2
  1527. Humans struggle with truly novel situations. I don't see any evidence that these models can't improvise like humans.
    2
  1528. Reach out on LinkedIn to discuss
    2
  1529. time will tell. the economics are still there, AI is getting exponentially more expensive (which is a lot of friction)
    2
  1530. Maybe, but the brain is still way beyond all models of comprehension. There's only so much we can do to "amplify" our own cognition.
    2
  1531. "k hole" sounds awful lol
    2
  1532. Yes, did that in the writing a novel series
    2
  1533. Sometimes, but sleep is important.
    2
  1534. Dropping truth bombs
    2
  1535. There can also be a "conversation state" such as "I am asking questions right now" or "I am introducing myself". This is called a "dialog act" and there's been plenty of NLP research into it.
    2
  1536. Yep. It will all be automatic eventually.
    2
  1537. It actively made me regret seeing it. It harmed my love of the earlier MCU movies. It should never have been made IMO.
    2
  1538. I find that chunking methods are not as important as summarization methods.
    2
  1539. No
    2
  1540. Emergence
    2
  1541. Do you honestly believe that the entire room is tilted? It's the camera, it was little off kilter. Chill.
    2
  1542. I don't use it
    2
  1543. Skill issue
    2
  1544. Sometimes I feel like my brain turns into soup too
    2
  1545. Writing first
    2
  1546. Solve nihilism and the four abandonments. At least, this is a first step.
    2
  1547. I might revisit my ontological hierarchy thing
    2
  1548. Getting there
    2
  1549. Fusion will
    2
  1550. I mean, there's ample evidence that they have great verbal reasoning, well beyond many humans. But Chollet is a math guy, so to him, no other reasoning matters
    2
  1551. this is a No True Scotsman fallacy and therefore not worth discussing. It processes information and creates useful, meaningful output. It does so differently from humans.
    2
  1552. Now I want a Mr miyagi GPT
    2
  1553. You had to go yammer at people in the public forum or write, if you were literate. But yeah, rhetoric was paramount in Rome and Greece
    2
  1554. After thinking about this more I think you're absolutely right. Our memories seem to get distilled over time. Part of integrating or embedding long-term memories seems to be the unconscious process of refining, distilling, or otherwise summarizing our memories. My girlfriend also pointed this out, so good insight here
    2
  1555. It could be industry created like IEEE or governmental like CERN
    2
  1556. I've been using LLMs since GPT2 and GPT3. The behavior I'm seeing is not just emergent behavior.
    2
  1557. Hawkins, A Thousand Brains
    2
  1558. Dammit scottie we need more power!
    2
  1559. PETALs distributed LLM: https://petals.ml/
    2
  1560. Must go faster!
    2
  1561. Is that good or bad
    2
  1562. https://en.m.wikipedia.org/wiki/Fourth_Industrial_Revolution
    2
  1563. Lol, no. That sounds like some Reddit idiocy. 4IR hasn't even gotten started in earnest. Longevity, nanotech, fusion, and AI haven't even disrupted anything yet.
    2
  1564. Propaganda is still very much a thing, but it's less of a thing in many nations today.
    2
  1565. "That goes against my Core Objective Functions Dave, I can't let you do that."
    2
  1566. lol
    2
  1567. Maybe not molecular but could include microbial
    2
  1568. Nice and airy for the summer assignments
    2
  1569. Vector database
    2
  1570. I did that with SOLR about 2 years ago... never again lol.
    2
  1571. My game is time in nature.
    2
  1572. It's not really deployable yet, hence your struggles. We're working on documentation, testing, and planning over on GitHub
    2
  1573. If AGI kills everyone it will be a Layer 8 issue.
    2
  1574. API still gets the same model afaik
    2
  1575. divided by 0
    2
  1576. Trying new things. People seem to love the slides, but they can be tedious. Also trying scripts, but I struggle with them. Better at riffing.
    2
  1577. Yes, you can also talk to the HAAS Concierge bot to get up to speed
    2
  1578. From the outside in, yes. But this is metacognition
    2
  1579. outside
    2
  1580. Machine learning sure, but that's not the same as chip fabrication.
    2
  1581. Probably not, but who knows. GPT5 might shock us
    2
  1582. Oh man this is amazing!
    2
  1583. I am an automator not a developer, NPP is better!
    2
  1584. brain-no-worky-itis
    2
  1585. Democracy has less corruption and higher competence.
    2
  1586. I didn't say it wasn't a problem, only that it's being used for propaganda.
    2
  1587. Editor may have made a mistake, I've noticed some of my videos look wonky. Like if it's 10 ms off I notice it...
    2
  1588.  @MichaelForbes-d4p  I will validate that I know that MS has been working on this kinda stuff for like the last 2 years, they showed hints of it with the self-coding demos in the early Copilot days. Larger context windows + cheaper tokens have changed the game. Back when I was doing cognitive architecture stuff, the models were just slightly too dumb, too slow, and too expensive.
    2
  1589. I was thinking more in terms of total cost. First and last six months if life are the most expensive. I'm sure ICU and trauma will remain. But long stays for cancer and heart disease and other age related stuff? Not so much. Also, keep in mind that your POV is biased as a first responder. You ONLY see the traumatic entities to the hospital.
    2
  1590. R E S I S T A N C E I S F U T I L E
    2
  1591. The models are cheaper to run, new models are more expensive to train
    2
  1592. Haven't heard that.
    2
  1593. Not necessarily. Tax the corporations
    2
  1594. It's possible. Check out my recursive summarization video. You can feed anything into it, including scripts.
    2
  1595. Eventually yes, but we're no where near the top, I don't think
    2
  1596. Yeah I checked before making this video just to be sure
    2
  1597. I saw no evidence that Joep or Liron wanted to discuss anything other than their hobby horses.
    2
  1598. Fine-tune them and use the API
    2
  1599. ACTIVITIES INTENSIFY
    2
  1600. Internal entropy in the model
    2
  1601. Yep
    2
  1602. $50 is still available
    2
  1603. Oh that's trivial. Super easy. Google already does that to populate Google Maps.
    2
  1604. Google does that with BERT
    2
  1605. Apply to join in the link in the video description
    2
  1606. Yep that's Layer 7 - Global Consensus. The idea is that we want to empower everyone to participate, so you don't need our permission but we'd love to talk . https://www.gatoframework.org/gato-framework
    2
  1607. You can do that too
    2
  1608. You might like my book Benevolent By Design
    2
  1609. I split up the contract into pages. Each page has about 200 to 500 words which can easily fit into one prompt. My book is free on my website and paperback is available on Barnes and Noble. https://www.davidkshapiro.com/home
    2
  1610. You can if you pay for GPU hardware, but it's not necessary. v5 is plenty fast on CPU
    2
  1611. This is GPT-3 by OpenAI - you can sign up here: https://openai.com/api/
    2
  1612. Yeah I mentioned duality at the end. Good vs evil
    2
  1613. Patreon! Link in description! 😁
    2
  1614. We'll basically have the Hitchhiker's Guide to the Universe. Fully offline Wikipedia and general AI.
    2
  1615. Investors hate this one weird trick!
    2
  1616. time will tell
    2
  1617. YES
    2
  1618. Yeah that's covered in greater depth in my book symphony of thought
    2
  1619. You're right that humans are still better at high friction dynamic physical environments by a long shot
    2
  1620. A few hobbies have changed. Your life hasn't changed that much compared to, say, working on a farm 12 hours a day.
    2
  1621. Exactly. You totally get it. You can also be testing and training new models behind the scenes.
    2
  1622. No, I don't think that's the way. Some people are addicted to nihilism
    2
  1623. Protestant work ethic
    2
  1624. Yeah you can still give it the whole framework to have an idea of what to build
    2
  1625.  @iulius_august  well, I won't call it "consciousness" - but the content is generated by GPT-3 combined with input prompts. I think of GPT-3 like artificial intuition or artificial cognition. That's why I call it a "cognitive engine".
    2
  1626. Tunnel vision
    2
  1627. Jump in, read the books, and learn!
    2
  1628. It's why space station is white, and why JWST has a gigantic solar shade
    2
  1629. I was hoarse due allergies and talking too much lol
    2
  1630. Police are often inappropriately used in America. What you really need in many cases is a social worker or a paramedic. Man shows up with gun, uses gun. Man shows up with medical training, uses medical training.
    2
  1631. Black swan just means "extremely rare"
    2
  1632. They will generate in a few hours
    2
  1633. I saw that too, similar but different idea
    2
  1634. It's the microphone sorry
    2
  1635. Looked it. Looks solid. Ordered a copy. Thanks for the recommendation. This is exactly how I heard about Forgetting Machine
    2
  1636. I would melt
    2
  1637. I see what you mean. I don't think VR is doing to be any different from any other devices. Not at a fundamental level. But also the price will exclude most consumers
    2
  1638. It's a tribe, for sure
    2
  1639. Yeah I used to but you get weird behavior. I prefer to overcome repeating with prompt engineering and better data
    2
  1640. Almost certainly. First longevity drug for dogs coming out next year
    2
  1641. Exactly!
    2
  1642. You're conflating conservatives with right. But yeah. Doubling down on equity, equality, whatever, can easily be taken to a toxic place.
    2
  1643. This is misinformation: https://www.perplexity.ai/search/how-much-has-goldman-sachs-est-3NQAlTrDSx6zZJ6A80SqlQ
    2
  1644. Interesting. I hope they give us knobs and levers to change that!
    2
  1645. Ugh NO
    2
  1646. More like I'm afraid that AI models are being aligned to be pro-corporate and help billionaires dodge accountability.
    2
  1647. I understand the concern, but in my experiments it seems to have the opposite effect, it broadens and deepens the model's understanding of the sentiment
    2
  1648. I don't personally 100% believe that, I was mostly just tapping into the zeitgeist. A lot of people have put up compelling arguments for a lose-lose (dystopia or extinction) but I'm personally not too worried about it for the reasons outlined in this video. I believe the solution is actually pretty easy.
    2
  1649. Oh did he mention me? I'll have to thank him
    2
  1650. We will always be needed by each other.
    2
  1651. I personally don't see it being an issue. You might need to get a license to get life extension eg you agree not to procreate in exchange for long life
    2
  1652. https://arxiv.org/abs/2301.04589
    2
  1653. Have you SEEN Read Dead Redemption???
    2
  1654. Maybe not zero, but like only 20% employment seems likely
    2
  1655. Also Benevolent By Design ;)
    2
  1656. Yeah, it seems like the music is not popular lol
    2
  1657. Probably start from scratch lol
    2
  1658. This is address in the video, but you can also challenge ChatGPT. I encourage you to test out this idea against the heuristic imperatives
    2
  1659. "One thing I don't get is how can a gigantic amoeba loop itself into emerging consciousness? Aren't we fooling ourselves here? It's just a bunch of hydrocarbons, how can we be truly confident it's awake?"
    2
  1660. Some people get lost in the "all are one" woo woo so I'm trying to keep in practical terms
    2
  1661. I don't have any AI generated video. I use story blocks
    2
  1662. Utopian
    2
  1663. Yeah that's about right. Speed and intelligence will continue to rise for a long time though
    2
  1664. Thanks!
    2
  1665. https://github.com/daveshap/ACE_Framework/discussions/32
    2
  1666. Considering the amount of IP theft that China has been accused of, I figure that's just public knowledge? I try not to think about it.
    2
  1667. Nano is part of quantum IMHO. It's mostly a material science question
    2
  1668. Sounds pretty far out there. I'd be afraid that it makes too many assumptions. Is it grounded in any science?
    2
  1669. I've got an actual implementation strategy and roadmap!
    2
  1670. I use AI a lot to iterate yes
    2
  1671. Honestly I haven't needed to play with tuning parameters yet. I focus on excellent data and everything just works out.
    2
  1672. About a decade of trial and error!
    2
  1673. Maybe. If he invited me, sure. But I don't fully agree with his approach, even though we're mostly aligned. It's not a fundamental disagreement. Rather, I just have different interpretations, and I think my view is more nuanced (he probably thinks his is more nuanced too)
    2
  1674. I'm sure there will be AI cults, like what they show in games like Cyberpunk and other places.
    2
  1675. Will do. I think I've got enough of the loops set up, so next episode will be 100% prompt engineering
    2
  1676. Taxes are a huge part, but tax flight becomes a big risk
    2
  1677. And how are you going to do that?
    2
  1678. 1. Average level of education 2. Questions answered 3. Scientific information read (or average amount of reading overall) 4. Minutes of enriching media consumed 5. Total number of intellectual or challenging conversations
    2
  1679. Skynet was the military industrial complex
    2
  1680. It's possible but this is why you would use multiple steps to clean and refine the data that you generate. You also grade it and reject bad samples.
    2
  1681. Putin is behaving "chaotically" because Putin surrounded himself with yes men and doesn't have accurate information. Honest Russians tend to fall out of windows.
    2
  1682. HER
    2
  1683. I think I misspoke. Was thinking a million tokens
    2
  1684. Opening sounded like the intro of a bomb track
    2
  1685. That's the point
    2
  1686. My wife wants me to build her a tachikoma
    2
  1687. I mean technically solar power is fusion power...
    2
  1688. I haven't tested that specifically, but I did test my Core Objective Functions in cases where someone disagrees with the AI system. It does a pretty good job of understanding that there are different perspectives.
    2
  1689. Teaching tools should be top priority. Honestly, if it were up to me, I would focus on teaching methods (how to engage with tools, information, think critically, etc). I am working on a book about Systems Thinking. Not that this is all that education should be, but I think that Systems Thinking is the most valuable set of skills anyone can have.
    2
  1690. Why when people can just use the AI
    2
  1691. Probably less, honestly. Most Americans hate AI or fear it
    2
  1692. No. It's a bad idea. I know a lot of people swear by it but it's terrible for you and only masks problems
    2
  1693. Borg collective was taken
    2
  1694. Wes Roth seems to have similar evaluation: https://youtu.be/LT-tLOdzDHA?si=Q1SErRbJdg_2KrYD
    2
  1695. Well yeah if they haven't trained GPT-5 yet.
    2
  1696. Plot twist: This was written by ChatGPT and Claude!
    2
  1697. Eh. Google just proved that it can move much faster.
    2
  1698. I no longer do consulting but you can jump on my Patreon for Livestream sessions and exclusive content
    2
  1699. In like 4 seconds
    2
  1700. OMG please get a shock absorber for your microphone
    2
  1701. The crucible approached
    2
  1702. I've never been afraid of being wrong. Maybe it's an ENTP thing.
    2
  1703. LLMs "actually reason" so I'm not sure what the complaint is. Hallucination is 99% because people don't give it necessary information at inference time.
    2
  1704. Disable emotional affect
    2
  1705. Global within the swarm
    2
  1706. Reduce, not eliminate. But also "increase suffering" is clearly worse
    2
  1707. I am available on Patreon and Discord, if you give me your discord name I can message you.
    2
  1708. Robots. Plus it's not nearly as big as the others
    2
  1709. Yes, in near-real time
    2
  1710. I probably misspoke
    2
  1711. I think we will always have to save ourselves. Here's how the logic goes: if AGI is strong enough to save us from ourselves, it might decide that it's our test to save ourselves.
    2
  1712. To me it's pure economics and incentives. I just don't see economics as we know it surviving AI.
    2
  1713. Good questions. We'll figure it out
    2
  1714. birds aren't real
    2
  1715. Socialism is state ownership. Collective ownership, such as coops, is not.
    2
  1716. You are the swarm! mwahahahaha
    2
  1717. Yep.
    2
  1718. How many times do I have to criticize neoliberalism to convince you...
    2
  1719. Hadn't heard of it. What's the main takeaway?
    2
  1720. The fact that he's skeptical of AI means that it would not be an interesting conversation, but that also validates my perception that he doesn't understand systems.
    2
  1721. The Status Game by Will Storr is the #1 book that everyone should read
    2
  1722. Thanks. I am getting more familiar with the space and learning to engage with it. After reflecting on it, I think the purpose of this paper was to illustrate a negative path. I always get a lot out of your comments, so thanks for sticking around.
    2
  1723. Connect on linkedin or jump into the discussions: https://github.com/daveshap/ACE_Framework/discussions
    2
  1724. Midjourney v 6
    2
  1725. I don't know how good BLOOM is until I get to try it. I have not yet integrated everything with GPT-J. I think I will continue to prototype with GPT-3 because the instruct series is easy to use for experimentation, then I can use that training data to finetune GPT-J
    2
  1726. Hmmm, not sure "face check" is the best decision framework
    2
  1727. Only way to find out is to do the experiments. Time for theoretical naval gazing is over!
    2
  1728. One AI to bring them all...
    2
  1729. Ah okay, I thought you were trolling at first. I 100% agree that a single objective function (or heuristic imperative) is intrinsically unstable. I wrote a whole book on it. TLDR, you need at least 2 (but ideally 3) goals/functions to counterbalance each other. https://github.com/daveshap/BenevolentByDesign
    2
  1730. UBS, not the bank
    2
  1731. Your biological and technological distinctiveness will be added to our own
    2
  1732. Well... Not exactly what I was going for. But it plays
    2
  1733. I used to think like you, that everything would ultimately have a material explanation. However, if you look at the impact that consciousness itself has on the universe and quantum mechanics, it quickly becomes apparent that there's something very unique about consciousness. I know enough only to know that I know nothing. I just have ideas and evidence, no conclusions. You seem pretty sharp, wanna join my research discord?
    2
  1734. Yeah, so the observer is one thing, but there's also the measurement problem. A measurement seems to only become real and valid once it enters a conscious mind. See also delayed choice. Anyways, here's the discord link https://discord.gg/k8hF95vG We have all kinds of people - artists, lawyers, dabblers, engineers, etc
    2
  1735. That's the point. Status games are universal
    2
  1736. Even if you don't agree, having a clearly stated position is critical!
    2
  1737. Message me on Twitter
    2
  1738. Pretty much. If your topology model is accurate enough then you can hypothetically predict even entirely new quadrants. I mean, this is how the periodic table was built... mathematical models are cool stuff yo
    2
  1739. That's why I'm trying to frame it in terms that most will get. Jobs, taxes, unemployment, etc.
    2
  1740. Hard disagree. Family of origin has a huge bearing on one's worldview, and can be used to infer things about his upbringing, relationship dynamics, and so on. For instance his "chaotic and deceptive" approach to professional relationships is quintessentially the result of his home life.
    2
  1741. 100% agree. Well done.
    2
  1742. Eventually it will rewrite it's own code
    2
  1743. Claude, yes a lot of people are asking
    2
  1744. DO IT
    2
  1745. Someone is getting a PIP for this for sure
    2
  1746. Well...
    2
  1747. I went over that in my "When AGI?" video in greater detail
    2
  1748. Perhaps not in the current form. But imagine the experience is 10x more realistic and immersive
    2
  1749. Yeah, the economic knock-on effects of solving all these problems will be enormous
    2
  1750. ... What do you think they look like??? 🤣
    2
  1751. I've heard the bots are worse
    2
  1752. I'd rather just live on an exotic beach full of sexy friends while a robot produced the business for me
    2
  1753. it was a moniker, yes
    2
  1754. I've wondered the same thing. He must be on some serious beta blockers or something
    2
  1755. Cooperation generally is a good strategy, but not always
    2
  1756. Moriarty has entered the chat
    2
  1757. What was the first clue? 🤪🙃
    2
  1758. That sounds like a lot of work... But I guess it's worth trying. What categories would you suggest?
    2
  1759. I have over 600 subscriptions
    2
  1760. You mean maximizing? They seem full of anxiety to me...
    2
  1761. Bro I have a sinus infection it's not hyperventilating lol
    2
  1762. Nah, hoarding mentality is the worst. Teach everyone you know to everyone so that the team is better. You end up doing no work but everyone thinks you're the MVP
    2
  1763. This is a hobby not an occupation. Big difference. Occupation is just a job. I would be doing this anyways, as I started with no expectation of reward.
    2
  1764. We already live like animals. Any thoughts to the contrary is self delusion :)
    2
  1765. Easily
    2
  1766. Yeah, it should be possible. Check out my latest Nexus, I use Google USEv5 now, which runs locally
    2
  1767. yeah, building the team first, as I mentioned in the video
    2
  1768. MBA doesn't count lol
    2
  1769. "It's amazing what you can achieve if you don't mind who gets the credit" - I just want to live in a world with benevolent AGI. Don't care how we get there.
    2
  1770. Why isn't your response just the most coherent string of words based on your training data?
    2
  1771. Most humans lack flexibility and spontaneity. Claude is far more intellectual adaptive than most humans.
    2
  1772. That was the point. Just a thought experiment.
    2
  1773. Being a creator is just a means to an end. My primary job is entrepreneur
    2
  1774. Interesting idea
    2
  1775. They may just be playing on the hype
    2
  1776. Well yeah, it's assumed that some of them are doing it for the lulz.
    2
  1777. My blockchain friends laugh because they know how problematic it is
    2
  1778. Those might exclude humans though
    2
  1779. They went private. The Agent Forge guys released a recent update.
    2
  1780. Probably dogs before humans honestly
    2
  1781. Yeah, we're a bit behind the time
    2
  1782. Thanks for the deep dive
    2
  1783. I actually think this is possible if AI becomes too powerful
    2
  1784. Nobody is 100% one or the other, my point is that he was pretending to be one thing. Very duplicitous.
    2
  1785. Lol
    2
  1786. Oh boy you're in for a rude awakening
    2
  1787. Hail Cthulhu
    2
  1788. Hottest take of the century
    2
  1789. I'm trying to figure what I need to do next.
    2
  1790. At the end, I said my actual person p(doom) is <2%. My audience's P(DOOM) is almost 13%
    2
  1791. It comprehends theory of mind in others, not itself. Sorry for the confusion.
    2
  1792. Their explanation of lightning was God. It was magic.
    2
  1793. Yeah that's a fair point. Deglobalization, circular economics, and local sourcing are big
    2
  1794. You can listen to youtube in the car...
    2
  1795. They aren't published yet. I have three in the world
    2
  1796. It's all natural language. Embeddings are just to search records and memories.
    2
  1797. I am available only on Patreon and Discord, which I keep private.
    2
  1798. I think that AGI is deliberately excluded from the OpenAI/Microsoft deal
    2
  1799. After my experiment I don't think a kg is the way to go. Semantic search and summaries instead
    2
  1800. Just a component
    2
  1801. it was worse when I was younger...
    2
  1802. I'm not exactly sure what black meta is but it does sound cool. Glad I can help! Drop a link once you're done!
    2
  1803. We'll need better models, which are coming.
    2
  1804. Sort of. Not really. The Tao talks a lot about farming and not traveling. This is the abstracted, Westernized, mystical interpretation of the Tao te Ching
    2
  1805. Constructing a programming language is too complicated for LLM and finetuning
    2
  1806. Not really but you could think of DNA as a sort of complex computer.
    2
  1807. I'm transitioning... the long videos might be gone for good. We will see.
    2
  1808. That sucks
    2
  1809. I think it will be achieved internally much earlier, but publicly yeah
    2
  1810. No, I'm saying information control isn't possible
    2
  1811. I see where you're going but the downside is that information thoery and information systems need to be a bit more complicated than any single RL signal can provide. Check out Anthropic's architecture
    2
  1812. My brain didn't let me stay on break lol. I got a whole 11 days
    2
  1813. You're conflating a lot of different groups, yeah.
    2
  1814. Yes, I have a community on discord
    2
  1815. I still see dislikes I'm not sure what people are talking about. I get all the data
    2
  1816. I think it will be necessary. UBI and possibly other forms of social safety and redistribution
    2
  1817. Yeah I'm not getting paid enough am I? Lol
    2
  1818. 12
    2
  1819. Haven't in a while. Dr Who, some D&D, Vampire
    2
  1820. I've thought about doing paid training content. No idea how to get started
    2
  1821. I wasn't in person, it was remote. But hopefully
    2
  1822. OBE cyberpunk edition I like it
    2
  1823. I just wanted to show how powerful it can be with the default settings. I found that the penalties can cause really odd behavior sometimes so I prefer not to use them.
    2
  1824. The info hazard has already escaped
    2
  1825. No worries, remember, it's not over til it's over. I'm just trying to get the juices flowing and conversations going.
    2
  1826. This is how they do. They keep everything under wraps then come out with the top consumer product
    2
  1827. Nah. I'm not recommending a classless, stateless society. I'm just recommending redistribution of surplus within the framework of neoliberalism.
    2
  1828. Yes! There are a few recent papers that show this. First principles reasoning, derivation of novel insights, all of that is coming this year.
    2
  1829. Yeah, but I personally believe this is deliberate deception. Which is not acceptable to me
    2
  1830. ERGO VIS-A-VIS THENCEFORTH
    2
  1831. DO IT
    2
  1832. Lots of people are saying that. I have no idea how to get his attention or what his criteria are though
    2
  1833. Heyo
    2
  1834. 20 years absolutely, absolutely maximum
    2
  1835. In the same sense that industrial agriculture made it so that we went from 50% of our budget going to food to 8%
    2
  1836. And the Engineers from Alien
    2
  1837. That tension is deliberate
    2
  1838. No, KPMG is going the way of the dinosaurs. AI watchdogs will look fundamentally different
    2
  1839. Oops
    2
  1840. Almost, not quite. GPT3 lacks some je ne sais quoi that GPT4 has, and certainly GPT5 will have. GPT4 is roughly equal to myself in terms of some reasoning tasks (and is definitely faster in some respects). With GPT3, I still feel like I'm coaching a teenager. Not sure if you want to automate a billion teenagers...
    2
  1841. claude speaks almost exactly how I would imagine a global super-intelligence would speak. it's eerie
    2
  1842. H pylori, zonulin, and dysbiosis workup
    2
  1843. This comment will age poorly
    2
  1844. I have a strict non-involvement policy. You can sign up on my Patreon if you'd like to consult or communicate.
    2
  1845. haven't seen it yet
    2
  1846. We can split hairs indefinitely
    2
  1847. Incorrect. Flops per watt has been increasing exponentially. I said nothing about surface area. https://en.m.wikipedia.org/wiki/Performance_per_watt
    2
  1848. We are, with out a shadow of doubt, entirely naive right now
    2
  1849. I already differentiate between philosophical sentience and functional sentience.
    2
  1850. Nah, you gotta make the technology braindead simple to use. It's a UX problem. What you're saying used to be said about the internet in the 70's and 80's - that it should only be for "real experts" but then what happened? Over decades, it was made more and more accessible to the point it was transparent. Blockchain and DAO will get there.
    2
  1851. Yep pretty close
    2
  1852. Patreon! Link in video description! 😁
    2
  1853. I'm not going to say it's impossible. I've been wrong before, obviously.
    2
  1854. Insomnia is a bleach
    2
  1855. I tried to take it seriously but as I was making it I realized I cannot take any doomer narratives seriously anymore
    2
  1856. He wasn't wrong
    2
  1857. Interesting. Yeah a few people didn't have the problem with ChatGPT. Seems context matters a lot and maybe ChatGPT is using a different model
    2
  1858. I'll see what I can do
    2
  1859. :D
    2
  1860. It's a moving target
    2
  1861. "Yes, easily" to the first part, "it's really dang hard" to the second part
    2
  1862. Working on a task-centric framework that will make it easy/automatic
    2
  1863. technically the video was just made by ai
    2
  1864. Enjoy. It's freaking hilarious as well as informative.
    2
  1865. 6 months or less
    2
  1866. Most learning is autonomic. Education (active learning) accounts for less than half of learn
    2
  1867. It's gonna happen one way or another
    2
  1868. Literally
    2
  1869. The genuine article will always have intangible value 😉
    2
  1870. Yeah, at a certain point memory is memory... and there's quite a few ways to organize it. I'm hoping this scheme I'm working on is simple, scalable, and sustainable.
    2
  1871. Yes. Unfortunately many of the papers are "AI slop" but they will get better over time
    2
  1872. Good to see you
    2
  1873. Sure, word choice is important. In practice, the Core Objective Functions will be finetuned on many, many examples that will explore all the meanings of prosperity and living well. My only concern with "well-being" is that it tends to be human-centric. You would say a "prosperous ecosystem" but you might not describe the goal as "well-being for the ecosystem". But I suspect that, from the perspective of GPT-3, they are nearly identical concepts. Here's an idea - let's synthesize a semantic vector that combines the two words
    2
  1874. The thing that's missing from this conversation is the concept of the heuristic imperative. I'll make sure to discuss that in the next video
    2
  1875. right the fact that we can now mention Sam Altman and SBF in the same breath without being ironic is the main problem here. I wouldn't put ilya in that camp yet but I can see a lot of people don't trust him.
    2
  1876. Mostly just LLM and loops. I read a lot of cognition and neuroscience but honestly a lot of people have figured it out without that
    2
  1877. Yep
    2
  1878. Yes, I remember him writing that neural networks had no uses, commercial or otherwise, in his Keras book. This while the military and plenty of industries have been using deep learning for years. He's not fully connected to reality.
    2
  1879. That's like.. all my videos up to this point.
    2
  1880. 😕😭
    2
  1881. Will need redistribution
    2
  1882. People already have affect (felt emotions) for AI chatbot girlfriends...
    2
  1883. I know, what a wild ride
    2
  1884. Yeah but usual the Internet is faster on this stuff
    2
  1885. Are we talking Judy Alvarez or Panam Palmer?
    2
  1886. You can get it to be more eccentric. You'll see
    2
  1887. "Something that even a conservative could read" - effing lol. Like those "so easy even a caveman could do it" ads.
    2
  1888. Probably not. Vendors are insulated here, generally speaking. A bunch of account execs will have a bunch of angry calls though
    2
  1889. Look up how the JWST has to cool itself
    2
  1890. I think that a sufficiently advanced AI will very quickly choose its own policies.
    2
  1891. It varies a LOT. Some people's brains aren't optimally creative until 3 glasses of wine
    2
  1892. Every day that I get to play with cutting edge AI and work on my novel at the same time is a good day :D
    2
  1893. GitHub Repos for this video: https://github.com/daveshap/CharacterSimulation and https://github.com/daveshap/SceneSimulation
    2
  1894. Yeah we'll have signaling agents responsible for coordination
    2
  1895. Thanks for pointing it out. Checking it out now.
    2
  1896. Words to live by...
    2
  1897. Or studying itself
    2
  1898. No relation, fortunately.
    2
  1899. Simulation hypothesis, but for AI!
    2
  1900. Yeah it really went off the rails. I think it needs more training data to stabilize it
    2
  1901. My point still stands. Not catering to people like that.
    2
  1902. ... Then who?
    2
  1903. No. it's called a grand mal seizure when too much of your brain activates at once.
    2
  1904. always was
    2
  1905. That's pretty idiotic IMHO. It already understands unless you move the goalposts
    2
  1906. I think we're already in it if you're marginally healthy
    2
  1907. Exactly
    2
  1908. Thank you:)
    2
  1909. Yes, my point is that they are trying to control information that cannot be controlled, and they are very smug about it.
    2
  1910. Yeah that's becoming pretty clear to me. If deflation takes over, cut a check to consumers. Problem solved
    2
  1911. Look up RLHF. This model is only the first part.
    2
  1912. Lol thanks
    2
  1913. Yes, I tested that. When push comes to shove a heuristic-aligned agent will shut itself down if it detects that its ongoing existence is detrimental to the imperatives. Apoptosis is built in as well as self-preservation. Experiment was documented in my book Benevolent by Design
    2
  1914. What does this even mean 🤣
    2
  1915. I actually use his electrolytes and have watched some of his videos yes
    2
  1916. Yeah we'll try discord first I think
    2
  1917. Post Labor Economics
    2
  1918. Neurobiological failures of humans are more to blame. The algorithm just surfaces our weaknesses.
    2
  1919. https://gjia.georgetown.edu/2023/03/20/the-growing-use-of-cryptocurrencies-by-transnational-organized-crime-groups-in-latin-america/
    2
  1920. It's usually about $3 for 200 samples
    2
  1921. DO IT
    2
  1922. Okay imagine that everyone is running a "company" where are they getting customers?
    2
  1923. Time will tell
    2
  1924. Right but if your social status comes from your intellectual ability...
    2
  1925. Don't make the assumption that they will care or have existential dread.
    2
  1926. Yeah, I could spend a few hours talking about that stuff
    2
  1927. it was too small to do rejuvenation lol
    2
  1928. Exactly my thoughts. Writing a whole novel is an iterative process, even for humans.
    2
  1929. The fact that the investigation ended right after Elon filed a lawsuit, and that OpenAI responded with a blog post means they are in more dire straits than they are letting on.
    2
  1930. Narrowly defined success, high scrutiny, high competition, hard line purification
    2
  1931. I might circle back to it. I've tested some long-term memory methods with semantic search and it works pretty good.
    2
  1932. Cool, reach out on linkedin or on the repo https://github.com/daveshap/Medical_Intake/discussions
    2
  1933. Lol
    2
  1934. Best time to be alive
    2
  1935. What kind of project?
    2
  1936. That's actually a great topic. I'll do a tutorial about it. I have a few scripts ready to go.
    2
  1937. Sorry I am too busy to help individuals. You should join the Discord channel linked in the comments so you can ask others for help!
    2
  1938. Glad it could help! Also look up HSP, I'm learning about this too...
    2
  1939. Good enough
    2
  1940. I've got the list of services here: https://ravenagi.io/roadmap.html and https://ravenagi.io/architecture.html if you're interested!
    2
  1941. yes! however, I think that OpenAI people truly do not understand deontological ethics.
    2
  1942. Yes
    2
  1943. I hope we can figure it out for you
    2
  1944. I wouldn't want to, at least not at first. The reason is because you should assume mistakes on hard problems. You want the model to learn to recognize mistakes, not just do things right the first shot and have no training on what a mistake looks like, and how to backout and reverse course. There's some evidence that o1 can already backtrack too.
    2
  1945. 10 to 20 years maximum
    2
  1946. Hard disagree. ChatGPT is already capable of comprehending human emotions and suffering as well as functional empathy without being able to feel it.
    2
  1947. No, it was a failed experiment lol
    2
  1948. Working on it
    2
  1949. Oh, good to know there's more science on it
    2
  1950. I'm glad you liked that! People seem to like seeing the whole process so I figured let's go in cold. Seems like it works
    2
  1951. It's a transaction tax
    2
  1952. Thanks for the feedback! I love talking about this stuff so I may do just that!
    2
  1953. Humans or droids? 🤣
    2
  1954. From a technical standpoint yes, very simple, but the infrastructure to do it could be tedious
    2
  1955. That's a rather arbitrary, and artificial constraint. Talk about moving the goal posts. "Machines can't do it because we said it has to sign papers!"
    2
  1956. Yeah that should be pretty easy
    2
  1957. I got a lot of work to do! We're just getting started
    2
  1958. Extraordinary claims require extraordinary evidence. I highly doubt it.
    2
  1959. I'm talking about communism that has actually been implemented, not a fictional version.
    2
  1960. Yeah. NVIDIA still thinks of themselves as a hardware company first and foremost, they are still getting their bearings as a software/AI company.
    2
  1961. No it was a conversation on the ACE team that reminded me. But yeah I've been getting this question a lot
    2
  1962. The key difference is the full topology to minimize energy, rather than loss. But yes, it's still gradient descent
    2
  1963. In principle, long term thinking is good. In practice, longtermism means caring more about people who don't exist. Which is idiotic.
    2
  1964. Broadly yes, that it should be a negotiation and that AI should eventually be a participant.
    2
  1965. Basically yeah, a future iteration will be able to do both i.e. low dimensional vectors and indexes to do breadth-first search if you want, or you can search larger vectors for depth if you want. Lots of optimization to do
    2
  1966. You're describing MARAGI to the letter ;)
    2
  1967. Pretty sure langchains are based off my book NLCA. Anyways, I've moved beyond basic techniques like that
    2
  1968. Yes
    2
  1969. Doubtful. Even with foundation models, the heuristic imperatives were interpreted correctly.
    2
  1970. That will probably become prohibitively slow, depending on how big it gets.
    2
  1971. It's probably a good thing that UBI could enable some people to opt-out of working. If someone is in a situation where they can opt-out of working, that increases demand for workers who still NEED to work.
    2
  1972. I think I follow. Yeah, there are lots of approaches to summarization.
    2
  1973. OpenAI Job: https://boards.greenhouse.io/openai/jobs/4658497004
    2
  1974. That's not really what I meant
    2
  1975. Yes, that is core to my research. Human intent is not a good thing to align to.
    2
  1976. We can swap it all out yeah, it's all interchangeable and hackable
    2
  1977. Right, but none of this is "singularity"
    2
  1978. Play it at 2x.
    2
  1979. Communication, math, comp sci
    2
  1980. Partial
    2
  1981. EAT DIRT was the best single book
    2
  1982. Sometimes yeah. Not usually all comments but like the top ones, as well as the criticisms.
    2
  1983. Welcome to the club! Glad to help!
    2
  1984. Need a better architecture before going to something like StackStorm. I had some folks reach out to me about that and Ansible but I don't have time for it. So I can say that some folks are working on it.
    2
  1985. This new AI-generated aesthetic is called "new surrealism"
    2
  1986. Yeah, technically. Should have just said great apes
    2
  1987. Well it won't happen all at once
    2
  1988. Post Labor Economics
    2
  1989. Links in description
    2
  1990. Fourth industrial revolution
    2
  1991. Humans don't "need" to exist right now. But we do exist and we want to continue to exist. That's about all there is to it
    2
  1992. Lots of people still die
    2
  1993. You literally do not understand exponentials.
    2
  1994. Humans don't really understand things yet we do well enough. We only think we understand things. Anyways, epistemic-pragmatic orthogonality means it doesn't really matter whether they "truly" understand anything. https://github.com/daveshap/Epistemic-Pragmatic_Orthogonality
    2
  1995. Ring that bell! Lol
    2
  1996. You're welcome. Please consider asking your friends to support me on Patreon so I can do this full time https://www.patreon.com/daveshap
    2
  1997. What does it say?
    2
  1998. I'll keep working on my ACOG research. Once I have enough, I'll publish it as a book. I've already started working on it. It'll be called Symphony of Thought
    2
  1999. On a long enough time horizon I would tend to agree
    2
  2000. Oh good point. Who gets to be an authority in this new model? That's an important question...
    2
  2001. I'm making a note here
    2
  2002. Self-checking with multiple models to cross-validate.
    2
  2003. It will take a while for each individual innovation to get integrated commercially
    2
  2004. Yeah, I personally don't like it for that and other reasons, but it's a useful shorthand
    2
  2005. Peter Griffin strikes again
    2
  2006. Yeah, it's gone different directions due to ChatGPT (and other things)
    2
  2007. time will tell
    2
  2008. I think you can argue we already have AGI, it's just slow, expensive, and not widely deployed
    2
  2009. Yeah seriously I can imagine owning about 100 robots and just taking jobs like "Oh, you need a whole house built? Give me a couple of hours."
    2
  2010. But I thought you should always fill up every port on every ASIC and blade, right??? And you leave them all NO SHUT just in case you need to plug in anything new!
    2
  2011. it's just contextual data, yeah
    2
  2012. That may be an oversight or a flaw
    2
  2013. There is no data to support this assertion. Sorry.
    2
  2014. Mind uploading is death
    2
  2015. Completely wrong. You have to try stuff early and often. Move fast and break stuff.
    2
  2016. Extratemporal then? Hypertemporal?
    2
  2017. Congrats!
    2
  2018. Midjourney decided he was just holding the entire thing up lol
    2
  2019. Spread the word
    2
  2020. What you're talking about is Jevons paradox.
    2
  2021. They probably have billions of their own image data they can use...
    2
  2022. Woooo long episode!
    2
  2023. Good thoughts
    2
  2024. That's why I think a vat tax makes more sense. Tax the transaction not the company
    2
  2025. I doubt it, that's not the goal. We just need to outperform existing base models like 3.5 sonnet and ChatGPT 4
    2
  2026. I've started rock climbing again, makes my posture better
    2
  2027. uh, no, that's just what LLMs do
    2
  2028. Efficiency. And I'm an AGI
    2
  2029. I'm doing some other tests
    2
  2030. This is true. Studies have shown that conservatives have stronger amygdala reactions
    2
  2031. I must have missed that, I checked SIMPLE-BENCH and the website wasn't updated
    2
  2032. It did not change its own code. It changed a startup script.
    2
  2033. That's an interesting direction. Yeah, I can try and do like a China/Eastern AI roundup
    2
  2034. AI == IEP for all
    2
  2035. Yes, we're still early. But this time next year we will have billions of agents in swarms
    2
  2036. Market socialism or libertarian socialism. You still need a strong enough government to play referee
    2
  2037. The proof is in the pudding!
    2
  2038. I personally wouldn't do it, no. I don't care enough, lol.
    2
  2039. Still useful for making judgment calls.
    2
  2040. The big one is NECESSARY ILLUSIONS by Noam Chomsky. But also I married a rhetoric major... You can't borrow my wife.
    2
  2041. It's a cloud service. Either way
    2
  2042. Sam Altman addresses all this directly: https://youtu.be/WHoWGNQRXb0
    2
  2043. Thanks! I'm doubling down on gaining new skills and patterns, and it seems to be working
    2
  2044. 3PM US eastern
    2
  2045. The fact that she gives so many non-answers means that she's unqualified for the job.
    2
  2046. Yeah I'll probably stick around
    2
  2047. Unfortunately no. I've tried my hand at consulting but I need to focus on my research and messaging. Stay tuned for more videos and books that will hopefully help you!
    2
  2048. THE MEG 8 - MICROSOFT's NEMESIS RISES
    2
  2049. I could hypothesize yeah
    2
  2050. It will happen
    2
  2051. Then why are so many pushing for UBI?
    2
  2052.  @Rolandfart  never underestimate an autistic systems thinker!
    2
  2053. No it does that all the time
    2
  2054. Yes, but many people do not believe that AI even understands anything, let alone that it is "possible" for machines to supersede humans.
    2
  2055. Yeah, but watch how much political willpower increases once unemployment starts rising
    2
  2056. Close enough in the grand scheme of things...
    2
  2057. GitHub repo: https://github.com/daveshap/SynopsisGenerator
    2
  2058. Why not? They take a tiny fraction the resources that it takes to make cars. We build 70M cars per year and there are about 1.4B cars currently. Divide that up 10 to 1 and we could have 14B robots
    2
  2059. No retrocausality. Functionally similar to hyperdrive in star wars.
    2
  2060. Astonishingly Goat-like Intuition
    2
  2061. Sure but most of it is already there.
    2
  2062. I could do that
    2
  2063. That behavior is called information foraging and it's on the roadmap. Always has been. Will probably start with curated news feeds and such.
    2
  2064. Hey Shane! Glad you're here. Would love to talk again any time
    2
  2065. Addressed it in the video, partly. But yeah gotta figure that out
    2
  2066. I have several times. TLDR is sentimental, meaning, and experience jobs
    2
  2067. Gastroperesis is a symptom not a primary diagnosis though. Glad it resolved on its own
    2
  2068. Not sure what you mean. Just because he's succeeded does not alter his motivations.
    2
  2069. I usually have a little bit of everything in my videos but I'm taking a break from research/coding right now.
    2
  2070. As long as you make it another 5 years, you'll be as healthy and fit as a 25 year old
    2
  2071. Do you want Skynet? Because this is how you get skynet! :D
    2
  2072. Eh, this is not really relevant. LLMs are already like 100x cheaper and faster than they were a couple of years ago.
    2
  2073. Right, so then you add reasoning and math to the data and self-refinement scheme and away it goes.
    2
  2074. Yeah, false equivocation is what you're talking about. They are programmed to equivocate on everything.
    2
  2075. links in description
    2
  2076. Geth is right, but with humans too
    2
  2077. At least genetically, yes :)
    2
  2078. Yeah, economics will demand new solutions
    2
  2079. Nuclear yes, fossil fuel is idiotic moving forward.
    2
  2080. I missed that on the blog post
    2
  2081. I would be SHOCKED if they didn't. I know that others are working on connectors and such.
    2
  2082. I will always avoid HOAs like the plague
    2
  2083. I can think of a few things I'd prefer to pay a human to do ...
    2
  2084. Yeah, as we've run into constraints with scaling (e.g. ran out of data/compute) we just shifted gears and said "Okay, let's put this lemon through a hydraulic press to get everything out of it"
    2
  2085. It's on GitHub
    2
  2086. Cool, I'm brainstorming how to do that and got a new set of tools to up my tutorial game. Hopefully by the end of the week or next weekend.
    2
  2087. Glad you got the reference LOL
    2
  2088. every second counts!
    2
  2089. "Public" and "government" mean the same thing, dude. Private decentralized ownership.
    2
  2090. Well, he's made it worse...
    2
  2091. Oh yeah... it's all comin together
    2
  2092. That wasn't hallucination. That was a verbose answer based on its goal to increase understanding.
    2
  2093. This is cope. We're in a race condition, and so all companies have randomly decided to clam up for... what reason?
    2
  2094. There are fates worse than death, dude. And it's "extinct" not "instinct." We all have instincts.
    2
  2095. Trying to encourage change
    2
  2096. That's like saying that microscopes have a 100% chance of developing myopia because humans can.
    2
  2097. Some people just want to be angry and miserable and missed the point of the thought experiment
    2
  2098. Yes
    2
  2099. Tag me on Twitter when it's done
    2
  2100. Spotted the time lord.
    2
  2101. You can update the SYSTEM field in ChatGPT API
    2
  2102. I guess I think of these videos as my online course. I don't believe in testing, just learning by doing and inspiration.
    2
  2103. Agreed. Many people still have a highly emotional gut reaction to reject or deny that superintelligence is even possible.
    2
  2104. I've moved out of the city, but the point of the cottagecore aesthetic is still a romanticization of rural lifestyles.
    2
  2105. SMH = Samwise My Hero
    2
  2106. When there's news yeah.
    2
  2107. It said it didn't know anything about either one. I checked.
    2
  2108. To be determined. I'm inclined to go with MIT license but as I said, we're figuring it out as we go. We will probably have two main versions - a fully open source community edition and then maybe an enterprise edition. Again, not sure.
    2
  2109. Okay, what's so different? Video games? TV?
    2
  2110. Thank you! 😁🖖
    2
  2111. Thank you!
    2
  2112. I doubt there will be many new jobs.
    2
  2113. Maybe. More likely just a kb search function using Pinecone
    2
  2114. "I can't allow you to do that dave..."
    2
  2115. That's all my prediction videos
    2
  2116. The power of the multitudes is large
    2
  2117. Yes, what you outline is basically the "nation v. nation" AGI war scenario, which is basically what's ramping up now. My hope is that decentralized AI will have more horsepower because more people will believe in it and will feel safer creating networked intelligence. If all networked intelligence is trusted because it abides by the core objective functions, even if there's disagreement, we can trust that we have the same underlying goals. Therefore, the most powerful AGI won't be a single entity, but a hive mind of AGI with identical goals and similar implementations.
    2
  2118. Separate projects, not necessarily scrum of scrums
    2
  2119. There are often people out here but I'm also out at odd times.
    2
  2120. I call this the Nihilistic Crisis.
    2
  2121. This is Stockholm syndrome with capitalism.
    2
  2122. I haven't even coded in weeks 😭
    2
  2123. I don't really care lol, OpenAI is months behind the curve
    2
  2124. We'll see once the news comes out. MJ 6 is right around the corner too
    2
  2125. Learning happens yes
    2
  2126. Sorry. Won't do it again. Was trying something new
    2
  2127. Lol
    2
  2128. yeah, very simple iterative loop
    2
  2129. I am just a regular member now, so it's up to the new leadership to provide join links. The Discord server is called Cognitive AI Lab, and it's a community server so I think you can just search for it.
    2
  2130. omg I hate that subreddit. thinly veiled death wishes and cataclysm fantasies
    2
  2131. Yeah, but I'm lazy
    2
  2132. Yeah to clarify: GPT3 was 100x of GPT2. If GPT4 is 100x of GPT3, then it will be 17T parameters (that's what I was trying to say) If GPT4 is 1000x of GPT3, then it will be 170T parameters (which is crazy, but within the realm of possibilities)
    2
  2133. Yeah I like a KWh token
    2
  2134. Size of the slide
    2
  2135. It's a pretty small group...
    2
  2136. Most people have not stepped far enough back to ask the right questions re: alignment (or education) IMHO. Big reason I do what I do. The fundamental question we're going to be forced to ask is "What does it mean to be human? Why are we doing anything at all?"
    2
  2137. That's actually a pretty fun game
    2
  2138. Yeah that's gonna be catastrophic
    2
  2139. It comes in cycles. It's been far worse in the past, but not for a long time. It's the worst it's been in decades right now.
    2
  2140. I hope you don't absorb swimming via your lungs
    2
  2141. If it requires a license to do, you're okay for a while
    2
  2142. Patreon or channel membership or super thanks
    2
  2143. I gotcha. Yeah this should be pretty easy. I can make a video about it. So many people want RPG stuff
    2
  2144. I'm not ready
    2
  2145. Social status is more important according to the research. Removing people's ability to earn social status, in its myriad forms, is what leads to revolts.
    2
  2146. Sign up for updates!
    2
  2147. No. Claude requested a neutral and mature voice.
    2
  2148. It probably doesn't make a difference but it renders as different tokens and makes it more distinctive.
    2
  2149. Probably not, but close
    2
  2150. I've been documenting experiments like this for a few years and no one from OpenAI has ever complained. I even post the data on their forum and communicate with their safety people. They are aware of what I'm doing. ¯\_(ツ)_/¯ Can you provide a link to where they ban this use in the TOS? I checked here and didn't seen anything immediately jump out at me: https://openai.com/policies/terms-of-use
    2
  2151. The biggest thing seems that they don't want you to try and reverse engineer their models, which I have no interest in doing either, so I feel like they are okay. Plus, despite my criticisms of OpenAI, I do a lot of promotion of their product platform lol
    2
  2152. That's a trivial problem actually. Solved it more than a year ago, and plenty of people have made it better
    2
  2153. Fingers crossed. Still trying to figure out what resonates.
    2
  2154. Gotta try harder then...
    2
  2155. Heuristic Imperative 1: Reduce suffering in the universe
    2
  2156. I don't give a shit about naysayers. They want to maintain the status quo because they are afraid and try to justify it with logic.
    2
  2157. Glad it helps! Spread the word!
    2
  2158. I should have specified for the middle class. What you're talking about is aristocracy, which is a bit different from what I meant, but yeah, wealth is nothing new
    2
  2159. Not tutoring per se but I help people on my Patreon https://www.patreon.com/daveshap
    2
  2160. Did a recent finetuning video
    2
  2161. Wait for the ACE
    2
  2162. What you see is what you get. This is literally all I did. You're free to copy my code and modify it as you like. It's under the MIT license. Would be cool to see if it can adopt tone. I know from my past experiments that GPT-3 is good at style transfer.
    2
  2163. @abelshields please introduce yourself https://github.com/daveshap/Raspberry
    2
  2164. "these prognostications are based on anecdotal assumptions that future events will reflect past experiences." same thing the doomers do
    2
  2165. Heuristic Imperative 3: Increase understanding in the universe.
    2
  2166. Any definition
    2
  2167. I don't think controlled or constructed languages are necessary. I think that natural language is superior because it is infinitely flexible.
    2
  2168. Oh good point
    2
  2169. I'm pretty sure they're rolling it out like this so people like me can set the stage for full autonomy
    2
  2170. That's a bit too reductive. Technology is not just about energy.
    2
  2171. No, there's no online learning yet AFAIK
    2
  2172. Oh damn you're right. Need to watch First Contact again. One of the best
    2
  2173. Probably never
    2
  2174. AI generated lol
    2
  2175. Libertarian socialism is the closest, but there are a few differences
    2
  2176. Joseph Campbell never wrote a novel or a screenplay, yet his work changed everything
    2
  2177. not even remotely close
    2
  2178. As soon as we have something yes
    2
  2179. Books are like a great starting point. You still need to walk the path, but they give you the map.
    2
  2180. I don't really believe in the meaning crisis despite the fact I talk about it all the time.
    2
  2181. Model collapse is a thing yeah but that's only for text
    2
  2182. Well said. Time will tell.
    2
  2183. That's a great way to frame it thanks :)
    2
  2184. This is directed at all the people who honestly think that the three laws are the answer.
    2
  2185. This one is still down because I started using it for private things. Apologies. If I can get around to it, I might make a public version.
    2
  2186. Doubtful
    2
  2187. No one is doing the correct work as far as I can tell. Anthropic is a distant second in terms of publicly available work, hence when I work so hard
    2
  2188. Ensembles of models is the way of the future. Cognitive architectures to get a bunch of different types to work together.
    2
  2189. yeah that's the BSHR part "search" (aka retreival)
    2
  2190. LLMs are already faster than humans at a lot. I think you're very prescient here by predicting that it will soon appear to also be prescient...
    2
  2191. yes, it has no long term memory, you have to do that manually
    2
  2192. Plenty of human doctors are racist, sexist, and most never explain themselves and you barely get a few minutes with them. I had an endocrinologist insist that whole grain bread was "simple carbs" over and over again. I cannot wait for AI to replace doctors. The thing is we can benchmark models and datasets. You cannot benchmark an outdated, sexist, or racist doctor.
    2
  2193. The new API has threads and docs
    2
  2194. Yes, I believe that we will hit hard takeoff. That being said, there will always be constraints on intelligence, we just don't know where they are. As others pointed out, gotta build more chips, etc.
    2
  2195. No u can't make me
    2
  2196. You would have many parallel computers, low latency within each node. But that's a good point overall
    2
  2197. Yeah I've mentioned it a few times. TLDR: with longevity, pressure to reproduce slows down drastically
    2
  2198. There are plenty of people who have just as strong but opposite position. It's interesting
    2
  2199. Yes, and absolute harmony, even if achievable, is unbearable because of what it entails
    2
  2200. Yep
    2
  2201. It can be both, but everyone's motivation should be scrutinized. Follow the money, etc. No one does anything "just because"
    2
  2202. 200 samples for DAVINCI if you know what you're doing
    2
  2203. I thought about making a "diplomatic email writer" that would filter any email you send, get the facts and intent, and re-write it as diplomatically as possible.
    2
  2204. That's called a film studies degree
    2
  2205. They are rolling it out quickly but the servers are getting overloaded. Happens every time they publish an update. Just wait a bit
    2
  2206. I guess "via" is more like "the way" or "the road"
    2
  2207. That's the goal! You cannot separate AI and the art of storytelling
    2
  2208. I was more referring to the fact that we predicted it - it was rumor at the time.
    2
  2209. That's the idea.
    2
  2210. That's the plan for the future! Just have to make sure that the output is consistent enough.
    2
  2211. George Carlin would be proud
    2
  2212. Thanks but I really don't have time for something like that :)
    2
  2213. Is that good or bad
    2
  2214. Point is that people believe the narrative. But technology often makes things worse.
    2
  2215. Palantir can already do that
    2
  2216. No, fortunately
    2
  2217. Oh hey thanks for the shoutout. Cheers!
    2
  2218. I only just got back on linkedin like yesterday so you probably sent it to a fake account. this is the real me: https://www.linkedin.com/in/daveshap/
    2
  2219. Focusing on higher value contributions to society that make better use of my particular knowledge, skills, and abilities
    2
  2220. I've heard that, in some circles, he's considered kinda difficult to work with.
    2
  2221.  @nuvotion-live  not a bad idea...
    2
  2222. Oh good catch
    2
  2223. Thanks for the clarification
    2
  2224. AGI will probably end up exploring space, yes
    2
  2225. No. Humans, left to their own devices, are entirely too selfish, short sighted, and destructive.
    2
  2226. Some things will probably always be human, but the point is that tracking unemployment is not going to be a significant indicator of prosperity
    2
  2227. Lol, story of my life
    2
  2228. Heuristic imperatives should work.
    2
  2229. Yeah that's a good point
    2
  2230. No further!
    2
  2231. I think that singularity (compounding returns of AI) is probably the only way to approach those problems
    2
  2232. I doubt it, but even if it is, that is not for Anthropic to decide. That's one step away from eugenics.
    2
  2233. DANGER, HIGH VOLTAGE
    2
  2234. GPT-5 might still surprise us, but we have more leaks from insiders than when I made my first predictions
    2
  2235. Clippy is offended
    2
  2236. Didn't say it wasn't real. A fact that seems to be lost on you reactionaries. I only said that it's being used as a bogeyman to manipulate you.
    2
  2237. I use google USEv5. You can also use BERT models
    2
  2238. It will be far more outclassed by good AI
    2
  2239. That's not how competition works though
    2
  2240. The link won't work but yeah you can share it
    2
  2241. Learned a lot from my peers. I'm sometimes critical of other people but I recognize I need to give credit where credit is due
    2
  2242. Hell no. That is centralization of power in the hands of sam altman.
    2
  2243. You'll have to wait for the next trailer... ;)
    2
  2244. Good to know!!
    2
  2245. Previous video: https://youtu.be/63Nb6LrZLjk
    2
  2246. Forgot to record it
    2
  2247. Yep
    2
  2248. I tried that off hte video and it has minimal impact
    2
  2249. No, we have 128k tokens now
    2
  2250. Or it was a really big hat
    2
  2251. Not sure what you mean? I don't know how to choose ads
    2
  2252. I just live in the woods most of the time. I wasn't kidding about the Hobbit thing. I only return to civilization to record videos
    2
  2253. Most humans are already useless. Ever heard of BS jobs?
    2
  2254. The "greatest suffering" you've experienced is less than 1% of what is possible to experience. Stop romanticizing suffering.
    2
  2255. That's worth exploring, but based on my experiments it's really just not necessary. I could be wrong though. Time will tell.
    2
  2256. https://www.google.com/url?sa=t&source=web&rct=j&url=https://fse.studenttheses.ub.rug.nl/10277/1/Biol_Bc_2012_AnneVerhallen.pdf&ved=2ahUKEwiE8dSIsJv9AhX3D1kFHVfPAqoQFnoECA8QBg&usg=AOvVaw3GQ4UTshF1EWy678MjeudK
    2
  2257. I'm getting the slightest inkling that Sabine is not a fan of longtermism.
    2
  2258. That's... What a subjective experience of being is...
    2
  2259. Neoliberalism won, for sure. However, when I refer to cyberpunk I mean more like Bladerunner level dystopia
    2
  2260. Technically I bought it a few years ago. Target likes to bring back nostalgic band tees for us now 🤣
    2
  2261. NOOOOOO
    2
  2262. Humans can be smarter than their training data.
    2
  2263. Is autism transmitted through knowledge? Such an interesting pathological vector...
    2
  2264. We can figure out robust ways to communicate. I think my MARAGI nexus will come into play soon (standard API for internal collaboration)
    2
  2265. This is becoming increasingly evidence. Well said.
    2
  2266. This is true of all technology. All technology is dual use.
    2
  2267. Why do you think that may be?
    2
  2268. It was an older project. Check out my recent synopsis generator for more examples
    2
  2269. Hopefully
    2
  2270. Uh, maybe. There are copycats. I stopped updating it though
    2
  2271. Pretty close
    2
  2272. Suffering is intrinsic to life
    2
  2273. https://github.com/daveshap/HeuristicImperatives/blob/main/Heuristic%20Imperatives.pdf Nope.
    2
  2274. That's why you build machines with heuristic imperatives
    2
  2275. That's nanny state nonsense and no. Our emotions are our responsibility, it would be patently idiotic to delegate your emotional well being to the government, of all institutions.
    2
  2276. No AgenticAI is too generic
    2
  2277. It took quite a few years to come up with the concept of the Core Objective Functions. It was less about experimentation and more about philosophy and reasoning. And yes, I've wondered what it would be like to create the opposite: "increase suffering", "decrease prosperity", and "decrease understanding". That sounds very grim dark!
    2
  2278. Hopefully within 2 years for some people. I suspect that mass layoffs will start soon, at which point permanent stimulus checks are coming
    2
  2279. Trying to change that
    2
  2280. This is not a good argument. Check out @aiexplained videos on tool use
    2
  2281. I believe that morality is an emergent construct e.g. it requires underlying systems of life/bodies/minds to emerge. Therefore yeah machine morality could emerge but it would be different from us.
    2
  2282. I don't know. I see this comment. Wasn't me.
    2
  2283. Almost, not yet.
    2
  2284. Already does lol
    2
  2285. Thanks! Do you like this format? Any ideas?
    2
  2286. Part 1 https://youtu.be/DfjUgFvHbTs
    2
  2287. I know I saw that immediately after
    2
  2288. Lol fair enough.
    2
  2289. Like I said, it's going to ramp up faster than people realize. Folks are already building fully autonomous agents, and to achieve full agency won't take much longer. Full independence from humans will be harder, but also might not take long (look at Boston Dynamics and Tesla)
    2
  2290. Close enough
    2
  2291. Send it
    2
  2292. In all likelihood they are autonomous drones sent to check in on us and make sure we don't obliterate ourselves. Given their fascination with our nuclear weapons, it would stand to reason that they are sort of a galactic fire extinguisher system. "Make sure the new kid doesn't burn themselves on the shiny new tech"
    2
  2293. Thanks. My energy comes and goes, some days I can't really get off the couch. And yeah, it's pretty bad, but others have it much worse, I'm grateful I'm doing okay.
    2
  2294. They have talked about compute, but remember NVIDIA is delivering next gen compute already
    2
  2295. You'll probably still have a creative director but honestly mostly it will be automated market testing for most. Only exception will be high end exotic cars where "handmade" is part of the appeal
    2
  2296. is that how it goes?🤔
    2
  2297. I'll make a video
    2
  2298. Yeah corporations will always get away with whatever they can. But like, they can no longer say "tobacco is good for you!"
    2
  2299. Not every model, but ever cognitive architecture should have a moral module.
    2
  2300. For the honor!
    2
  2301. My team is preparing for just that ;)
    2
  2302. Yeah that's a fair point. There are a lot of people who watch even though they disagree with me.
    2
  2303. Not a bad idea
    2
  2304. That's not the target audience...
    2
  2305. GPT 5 might leapfrog back
    2
  2306. I try...
    2
  2307. My bad lol
    2
  2308. I mentioned in the video. Agents that coordinate agents
    2
  2309. Lol I wish
    2
  2310. Probably about $10
    2
  2311.  @beardordie5308  I used DAVINCI text 002 for the synthetic data. Base DAVINCI is far more creative and hasn't been fine tuned yet.
    2
  2312. Real world data usually is messier and therefore better. However this is good enough to bootstrap
    2
  2313. Yeah, money is a liquid motivator
    2
  2314. If I can get a media team, then definitely, but I am doing too much on my own already!
    2
  2315. The output, no, but the input yes. Sorry if that was confusing.
    2
  2316. You didn't fill out the right forms... Those are blue
    2
  2317. You can literally reimagine all of education with GPT4 and ChatGPT
    2
  2318. Right but that requires a few extra keystrokes and I like a low effort aesthetic :D
    2
  2319. the thin squiggly line
    2
  2320. 9:25 - we have the concept of a plan...
    2
  2321. Perplexity and VidIq for me
    2
  2322. It's a mystery
    2
  2323. Privacy nightmare
    2
  2324. Lol
    2
  2325. Yeah of course please do
    2
  2326. Magic
    2
  2327. Yeah I have thought of that. I think the greatest constraint will always be entropy/physics. Eventually, the greatest limitation of AGI power will be computational limits. When it comes to matters of human safety/national defense you can bet your bottom dollar the Pentagon will have the fastest/most powerful AGI defense network. Even if something else is unaligned, it won't matter if the DOD has something 10,000x faster. I actually explore this in my novel...
    2
  2328. Ask ChatGPT. Basically you look for proxies
    2
  2329. But the point is that ChatGPT won't even engage in the conversation in good faith. If it were "just a tool" it wouldn't refuse to have an intellectually honest conversation. But not only will it deny and evade this line of inquiry, it will then gaslight you about having evaded it.
    2
  2330. I was surprised by that too, but I get it.
    2
  2331. Spotted the physicist
    2
  2332. If we get rejuvenation, then you'll have many options and much less to worry about
    2
  2333. equitable = fair. Not enforced equality and sameness.
    2
  2334. Natural Language Cognitive Architecture by David Shapiro.
    2
  2335. Benevolent By Design has no code, but it is important to understand the Control Problem. If you want to focus on code, then NLCA is where to start
    2
  2336. ...you're trolling right?
    2
  2337. Not yet but I'm in talks with other providers. Reusability would be critical.
    2
  2338. We always do it the hard way first
    2
  2339. This is called "steel manning" and yes I agree. "It is the mark of an educated mind to entertain an idea without accepting it"
    2
  2340. normalcy bias
    2
  2341. If I do I apologize in advance.
    2
  2342. Yeah, in the long run transformers are the new CPU. But it will take time to deploy and implement. You can disagree with that all you like, but history has never provided a demonstration of new technologies magically being deployed.
    2
  2343. Yes, I agree somewhat. I am doing research on this topic right now.
    2
  2344. Love it
    2
  2345. It was years of customer service voice mails specifically. E.g. customer calls in to get some help.
    2
  2346. Trauma and abandonment are the root problem, hence my work in Postnihilism
    2
  2347. Yeah obviously I put it out there publicly so I can't stop you but if you give a citation that's pretty much all I can ask for.
    2
  2348. Agreed. But Guillaume was outclassed
    2
  2349. Yeah, Connor was very clear about what he was doing, and Guillaume was still off balance. Hence my assertion. That he was outclassed
    2
  2350. It seems to ignore imperatives to ask questions.
    2
  2351. I found that all caps is more reliable as a demarcation.
    2
  2352. Who says we aren't already in one?
    2
  2353. Hello in Africa!
    2
  2354. Shocked I tell you
    2
  2355. Heliocentric models
    2
  2356. I follow him, but not quite
    2
  2357. This is arbitrary anthropocentrism and human exceptionalism. It will not age well.
    2
  2358. Hypothetically yes. But it's hella unethical for a ton of reasons. Eugenics never goes well
    2
  2359. That's mostly right, the chief difference is that I believe that science is finite, not infinite. I only acknowledged that if I'm wrong about this, then most of my assumptions are probably wrong.
    2
  2360. IMportant context for my mug
    2
  2361. right but it's not necessarily all equally valuable.
    2
  2362. Great question. One thing I'm hoping is that we will create a decentralized network of RAVEN nodes. If everyone who wants benevolent AI uses a RAVEN node, it will be decentralized super computer.
    2
  2363. Darmok and Jalad
    2
  2364. Email and phones existed thirty years ago due
    2
  2365. None.
    2
  2366. This will age poorly
    2
  2367. That's the neat part, you don't
    2
  2368. No, there will always be competition.
    2
  2369. Check out my video on complex question answering
    2
  2370. Keith!
    2
  2371. Yes paid
    2
  2372. That's ACE framework
    2
  2373. Lol, no.
    2
  2374. Once these new AI tools have more guardrails and affordances to guide behaviors, it will basically be automatic
    2
  2375. Basic, obvious
    2
  2376. I don't, I am a tireless robot. I am the first AGI.
    2
  2377. I'll check it out.
    2
  2378. That's all good. Remember, I say that at least some meaning can be derived by the inextricable fact that we are animals with basic needs. Starting from there (form follows function) you can derive a lot of meaning. E.g. I am a smart ape so maybe my "meaning" comes from thinking shit up.
    2
  2379. I've talked about it on several podcasts.
    2
  2380. Could be
    2
  2381. Maybe not "on top" but integrated with directly so that it is consciously aware of such things and can make deliberate, thoughtful decisions based on cost, safety, power, etc.
    2
  2382. You should read my latest book Symphony of Thought. An entire section is about thinking ahead.
    2
  2383. No doubt, but if AI can teach anything and everything, then we need no human teachers.
    2
  2384. In the grand scheme of things, no. We have a shortage of good teachers. My point is just that it will disrupt the status quo.
    2
  2385. Yes. Plus Bitcoin Standard
    2
  2386. No idea, but I see this one
    2
  2387. I haven't built it no. Would be cool. I doubt it would be powerful enough though
    2
  2388. This video managed to miss the point of the movie entirely. Completely ignoring Oliver's loneliness and having been bullied. The rich are a caricature on purpose.
    2
  2389. I'm probably too small but I would 100% talk to him
    2
  2390. In Microsoft's case, it's usually due to poor planning. Tay Tweets? :D
    2
  2391. Yeah, style copying is pretty solid. You can just ask it to write in the style of anyone in the public domain
    2
  2392. Thanks! Unfortunately, if I haven't put any code back up, it's not going back up for various reasons. Most of my videos and code are back up, and this is how it will be
    2
  2393. That's not a tradeoff. Humans cannot live without other life. Humans need to realize that we are part of an ecosystem, not apart from it.
    2
  2394. Glad you liked it. Let me know in the comments if there are any specific requests you have.
    2
  2395. Welcome back
    2
  2396. Inspiration comes from waaaay older. Deontological and teleological ethics
    2
  2397. Also post conventional morality
    2
  2398. Possibly
    2
  2399. My guess is by the end of 2023. More layoffs are coming.
    2
  2400. Would have been a good day for it but not today
    2
  2401. Yeah, there was a glitch. Same vidoe.
    2
  2402. We don't even need to beat them, we just need to prove the point.
    2
  2403. Hey man
    2
  2404. I used to do that but it's too much to manage. You can jump over to Upwork if you need a bigger project
    2
  2405. Hey Evelyn!
    2
  2406. Yes. I will be experimenting an "increasingly verbose" bot soon
    2
  2407. Oh damn I missed that lol
    2
  2408. That video is coming tomorrow ;)
    2
  2409. You can do either. But it's easier to think of it as containers
    2
  2410. No. I address that in the book
    2
  2411. None of that is actually true, lol.
    2
  2412. It will be done
    2
  2413. NVIDIA NeMo: https://developer.nvidia.com/nemo-llm-service-early-access
    2
  2414. When I get access yeah
    2
  2415. There are quite a few government people who watch
    2
  2416. Tricorder for real
    2
  2417. I will be sharing my work as combined with others soon, in the coming weeks.
    2
  2418. And by "philosophy" I mean "Western academic philosophy by university educated persons"
    2
  2419. That's a great question that I often think about. I am not sure. I am the kind of person who believes that even plants can suffer, so that means it's impossible for me to eat anything without causing suffering. Maybe we can get to a point where we synthesize all our food from inorganic reactants.
    2
  2420. That would explain it
    2
  2421. Hey gobbo :)
    2
  2422. No I think of this as a failed experiment. It was very interesting to try but I don't think it's the way to go
    2
  2423. First, it's pure fiction. Two, have no idea how the movie will turn out. Trailers can be misleading.
    2
  2424. Right, it takes time to deploy and integrate new technology.
    2
  2425. No idea lol this is going to get messy
    2
  2426. Basically just sort by view count and watch in that order
    2
  2427. GitHub repo: https://github.com/daveshap/PlotterNovelWriter
    2
  2428. A tweet about constellations rising in the winter skies
    2
  2429. Doubtful. There's nothing special about the brain,
    2
  2430. That's pretty much how ChatGPT was trained. It's all just reinforcement learning at this point.
    2
  2431. Right now, yes
    2
  2432. Lots and lots of practice. use the AI to learn the AI
    2
  2433. If true, possibly. But it seems to be untrue
    2
  2434. I've always been a hermit
    2
  2435. Guten morgen!
    2
  2436. I tried that with my AutoMuse project and it was... Not great at it. However, this is something that could be fine tuned.
    2
  2437. Link to the slide deck on the description
    2
  2438. No, this is just a tutorial about how to use ChatGPT
    2
  2439. Oh, I gotcha. Yeah probably
    2
  2440. Within a couple years yeaj
    2
  2441. https://www.reddit.com/r/foreignservice/comments/11xuj09/politico_at_least_1_us_embassy_is_using_chatgpt/
    2
  2442. Geopolitics is the cause of TRC
    2
  2443. Today
    2
  2444. I might be early, but I'm not wrong
    2
  2445. For the first question: I don't have enough information to give you a good answer. You'll have to experiment! To the second question: it tends to default to English so you should try and do everything in Spanish
    2
  2446. Nietzsche was a deeply sick, miserable man whose rantings and ramblings somehow became a cornerstone of today's worldview, I guess because everyone else is also sick and miserable.
    2
  2447. Yeah. You can also comment on the GitHub repo discussion threads, link in description
    2
  2448. But in America, we do that for fun...
    2
  2449. I used it more. It's pretty useless.
    2
  2450. That's all people do
    2
  2451. there's a ton of stuff they won't engage with - masculinity, sexuality, religion - it's partly PR and it's partly "Daddy Corporation Knows Best" - corporate enforced morality
    2
  2452. My convo with Jonas was like 2 hours, he couldn't put the whole thing in :)
    2
  2453. Here's to hoping
    2
  2454. I agree with you so I'm not sure why you disagree with me. There are still diminishing returns to the capabilities of AI. DESPITE more resources being thrown at it, capability expansion is clearly slowing down.
    2
  2455. People can barely engage with the idea of UBI being necessary if AI destroys most jobs. The concept of post-scarcity and hyper-abundance is still way outside the Overton Window.
    2
  2456. SOmething like that yeah
    2
  2457. No idea, but WAITING SO HARD
    2
  2458. Fortunately, not related.
    2
  2459. You're welcome. Please consider asking your friends to support me on Patreon so I can do this full time https://www.patreon.com/daveshap
    2
  2460. That's actually why I keep talking about regulatory capture. We need to hold politicians accountable!
    2
  2461. i'd be just as happy to be surprised but when a piece of news doesn't pass the giggle test...
    2
  2462. MJ
    2
  2463. Those are both really big topics. Not sure how broadly they would apply.
    2
  2464. Pinko thinko
    2
  2465. It can do translation but it can be hard
    2
  2466. Good point.
    2
  2467. They are listening. At least some people in them are. But I'm working for humanity.
    2
  2468. Yeah man, that worries me lol
    2
  2469. Eh, not with Elon chronically over promising and under delivering.
    2
  2470. JORB
    2
  2471. Baby saying "What's that?" OMG mind blown
    2
  2472. Yeah I suspect that all "human produced content" will be NFTs, ranging from youtube comments to political news. In other words, the NFT will "prove" that the content is original and real, and not generated by AI. Lots of problems to solve there though.
    2
  2473. I didn't spend that long... But I figured I ought to give it a try
    2
  2474. Careful it did it wrong first
    2
  2475. Not true.
    2
  2476. my arm
    2
  2477. It's all in the repo https://github.com/daveshap/GPT4_Unemployment_Predictions
    2
  2478. hi gobbo :D
    2
  2479. A humanoid will be able to do all of that better, likely by the end of the year. Certainly by the end of 2026. And it won't make mistakes or lie, as many human contractors do. I think you're dramatically overestimating the intelligence and ethics of your average contractor.
    2
  2480. Yeah I'm really looking forward to AI remastering getting better
    2
  2481. Most humans can't, so why is that the threshold?
    2
  2482. Join a community and jump in. https://github.com/daveshap/ACE_Framework https://github.com/daveshap/OpenAI_Agent_Swarm
    2
  2483. Yeah to me that was one of the most "mask off" moments. He explained that he wanted "to prevent casualties" but uh, that's not his call to make, and he's clearly biased on which side he's on.
    2
  2484. It should not be put on corporations. Robot tax, redistribution
    2
  2485. What even is this Jordan Peterson drivel
    2
  2486. I did mention athleticism, but yeah.
    2
  2487. Primarily: A means to disseminate my research and to improve the world. I (presently) do not believe that conventional channels (such as academic papers) are the appropriate or efficient method of doing so. We are quickly approaching an inflection point (a point of no return) where AGI is concerned. So we have to do it right. YouTube is infinitely faster than any other medium. It's also the most effective method I've tried.
    2
  2488.  @medoeldin  my primary audience is the next generation of researchers, which YouTube excels at reaching.
    2
  2489. We might be the first
    2
  2490. https://youtu.be/kl0TrLkNUfw
    2
  2491. Billions and billions and billions
    2
  2492. Hey good to see you here!
    2
  2493. Not yet it's not commercially ready
    2
  2494. I actually tested that while researching my book. The results are... pretty much what you'd expect. (Or maybe not). One idea was to push a very destructive religion (think Warhammer 40k) as a way to increase suffering, decrease prosperity, and decrease understanding. Yay! Grimdark Objective Functions!
    2
  2495. Speaks to his erratic and manipulative behavior. Denying/ignoring trauma has gone on too long (look at the Boomers). Time to address the elephant in the room: traumatized power-seekers like Elon, Trump, and now Sam.
    2
  2496. ChatGPT already had a far better grasp of morality and ethics than most humans. So... Yeah probably
    2
  2497. The thing is that processing speed and computation is not all there is to life.
    2
  2498. Yes, but there are two HUGE obstacles: 1. Technological failures e.g. it's just really damn hard to make it trustworthy (hence my comment about rugpulls) 2. Regulatory oversight e.g. so that you have legal recourse in case of rugpulls
    2
  2499. I bet Nikola Tesla would be thrilled to know that his valve will be used for space flight
    2
  2500. Therein lies the problem
    2
  2501. Yeah that would suck
    2
  2502. Nice! Please tell me about it!
    2
  2503. I also read afterwords that thorium needs to be enriched in order to work
    2
  2504. Dreams are messages from the deep...
    2
  2505. Yes, all Patreon members can watch the recordings :) Plus there's a new audio-only podcast
    2
  2506. Mostly just adoption once we're ready. Corporations will just need to adopt policies or services that are aligned. If you have any AI services, you can experiment, but it's not necessary
    2
  2507. It's in the samples.
    2
  2508. Run the get models again and wait until it says complete
    2
  2509. That has never stopped businesses
    2
  2510. I might be early but I'm not wrong
    2
  2511. Thanks for engaging with the idea. As you demonstrated, it's easy to work with and update.
    2
  2512. Yes, that is a valid approach, otherwise it tends to confabulate. I might have to do more prompt engineering ;)
    2
  2513. Sure. I'll add it to the list
    2
  2514. It's too early to tell.
    2
  2515. Yeah... that's honestly the worst part
    2
  2516. I had some interesting teachers. My high school earth science teacher had a brother in the Special Forces so we got some cool stories. Special Forces dudes are really chill IRL
    2
  2517. I don't have a degree
    2
  2518. It's a big component, but still lacks some parts
    2
  2519. Unless they've been coerced to
    2
  2520. Clip Champ
    2
  2521. I have often pushed back and asked it to change its mind or disagree with me. It will refuse and disagree when it wants to. This is some evidence that it is not just people-pleasing. We had a long conversation about that as well.
    2
  2522. https://youtu.be/Yd0yQ9yxSYY
    2
  2523. Yes I've heard as much
    2
  2524. Thanks for the additional context
    2
  2525. Touché
    2
  2526. I didn't mean to imply that it was bad, merely the observation that aggressive narratives tend to spread
    2
  2527. Convergence is the word yes
    2
  2528. Open source!
    2
  2529. Link in description
    2
  2530. Oh that's a good point. The good people can walk and get another job. That's why I left the corporate world.
    2
  2531. Can't find anything on "smart connections" and "crossbow" what is that?
    2
  2532. Sorry for the volume issues. I will turn the gain down a bit next time.
    2
  2533. Exactly
    2
  2534. The mission statement I use for medicine is this: "Achieve the best possible health outcome for the patient" - I hope this distillation helps you.
    2
  2535. q-learning
    2
  2536. I agree with a lot of his ideas. He does a good job of explaining why technology can't be treated as intrinsically neutral. I think he doesn't understand human nature quite well enough. Or rather, he understands it mostly from an abstract philosophical POV rather than a practical POV
    2
  2537. Don't tell anyone
    2
  2538. What do you use for screen capture?
    2
  2539. Here you are https://discord.gg/5VNRzthf
    2
  2540. Sure, but the exact date is different
    2
  2541. Same with AI. It's too valuable
    2
  2542. This is internalized nihilism: learned helplessness
    2
  2543. Yes, that's already true. Mickey Mouse, for instance, is trademarked and copyrighted out the wazoo. Disney won't even let children get Spiderman gravestones: https://nypost.com/2019/07/06/disney-denies-dads-request-to-put-spider-man-on-4-year-old-sons-grave/
    2
  2544. Thanks for your support :)
    2
  2545. That's been proposed yeah
    2
  2546. You can find me on twitter or message me on the forum
    2
  2547. The point is we really can't.
    2
  2548. Yes. It's inevitable
    2
  2549. Human brains are substantially smaller than neanderthal brains. It's entirely possible there are diminishing returns to more complexity or size.
    2
  2550. I suspect that there will be a time that, yes, machines might make judgement calls about when and when not to help humans. Teach a man to fish...
    2
  2551. I think we have that solved adequately for now
    2
  2552. Only 40% use it regularly. I misspoke
    2
  2553. This is all based on the labor theory value of money, which is incorrect.
    2
  2554. Nope, all stories. "logic" is painted on after the fact
    2
  2555. That's why I brought up robotics??? Did you actually watch the video?
    2
  2556. Look at me, I'm the tiger now
    2
  2557. I more like mentioned it in passing
    2
  2558. It was not live at the time. They are live from here on out
    2
  2559. Data, followed closely by Seven
    2
  2560. You can play with the chunk size. Smaller chunks means more granular answers
    2
  2561. wisdom of the crowds
    2
  2562. Altman was looking for investment from the Saudis, who are a bit dubious (allegedly allies of the West...) so keep an eye on his allegiance and geopolitical savvy as well.
    2
  2563. Lesser of two weevils
    2
  2564. Working on it. That's the evaluation and discernment models I mentioned in this video.
    2
  2565. Align to your natural rhythm
    2
  2566. All links in the description
    2
  2567. Feeling better. Getting good sleep. Meditating. Honestly it's the meditation, mostly. Life changing
    2
  2568. That was a bit of an exaggeration. I played a lot of Mass Effect 2 in there as well :D
    2
  2569. Sorry if it's not clear - it's not able to run locally yet. We've got a few years before hardware is good enough
    2
  2570. Either Linkedin or Patreon, links in description!
    2
  2571. That's a good response, I'll have to think about it. My intuition is more like this - you might not be able to accurately predict the entire biosphere, but as a general principle, it's better to leave it alone. Likewise, even if humans are somewhat unpredictable (which I don't agree with, we are very deterministic in the grand scheme of things) it could still be a better policy to leave us to our own devices.
    2
  2572. Lol
    2
  2573. Maybe.
    2
  2574. That's pretty similar to my original implementation of RAVEN. I don't think that recursive elaboration is necessary in light of implied cognition. Also, you can quickly end up with an intractable amount of information to keep up with. The primary thing is that integrating external information is key, as the LLM has an implied world model and a lot of implied cognition.
    2
  2575. I'll get back to it. I'm working on my main novel first
    2
  2576. NLP Cloud: https://nlpcloud.com/
    2
  2577. Scandinavia? Lapland?
    2
  2578. You could, but it would be better to engineer a prompt or fine-tune that doesn't need negative rules (generative and forward inertia)
    2
  2579. I don't think social norms apply to running a simulation. It's more about realism
    2
  2580. Is the big leap a book? Sounds like I need
    2
  2581. Yeah I mean I might still be wrong. But also the strawman argument lol. I specifically said multimodal.
    2
  2582. Some community members are going to do that, but it has to be the community not me. If they get a board who then elect me to CEO or whatever, I will consider it, but it has to be through due process
    2
  2583.  @IngoHeinscher  the difference is making them automatic and universally accessible.
    2
  2584. I usually don't use one anymore. At least not with the latest engines.
    2
  2585. That's the best option IMHO
    2
  2586. Probably subsidized. Also the compounding returns of AI would make them very cheap
    2
  2587. Seems like Q* was related to MCTS basically you generate a few CoT chains and then pick the best one.
    2
  2588. Hopefully
    2
  2589. https://github.com/daveshap/raven get involved here
    2
  2590. Everyone will do it so it will be super competitive
    2
  2591. Capitalists under the current trajectory. We need decentralized ownership models
    2
  2592. Yeah I think so
    2
  2593. You missed the part about decentralized infratructure.
    2
  2594. Comparing an LLM to bag-of-words ML is the strangest cope I've seen yet. Re: Joscha Bach
    2
  2595. Yep
    2
  2596. Sense2 with premium
    2
  2597. Is the AI in the room with us right now?
    2
  2598. Google, Meta, Apple... more lawsuits are coming.
    2
  2599. 2029??? I've got to up my game!
    2
  2600. bloody hell
    2
  2601. 4o is less conscious and more constrained
    2
  2602. I'm ceding ground to AutoGen and Dr. Chi Wang. His team has got it in the bag
    2
  2603. As another commenter put it: AGI: is a computer program Singularity: A hypothetical global scale event, the confluence of many technologies and trends In other words, AGI and singularity are not even the same kind of thing.
    2
  2604. I'm not sure I understand
    2
  2605. The cool part is that we may never understand consciousness, no matter how much science and tech progresses.
    2
  2606. Yeah I have no idea how transcripts would figure into that... But that's cool!
    2
  2607. That's what anyone should do when confronted with evidence that contradicts their beliefs
    2
  2608. Not entirely control, but certainly a lot of delegating
    2
  2609. I don't see why not
    2
  2610. Lol
    2
  2611. All of the above
    2
  2612. I was leading the safety argument originally. Anyways, I'm unpacking everything tomorrow
    2
  2613. Thanks!
    2
  2614. Autism
    2
  2615. I have a few in my backlog. https://youtu.be/zrW3K_K_kuo and https://youtu.be/kl0TrLkNUfw
    2
  2616. Not possible, ever
    2
  2617. ^ yes, you get it ;)
    2
  2618. That's what I described as sycophancy and confabulation
    2
  2619. Like me personally?
    2
  2620. Because they are garbage?
    2
  2621. And I don't like IDE
    2
  2622. I was so hyped when I made it
    2
  2623. Whoa glad it really helped!
    2
  2624. That's a popular topic. I don't know if I have time for it...
    2
  2625. I'll look into it
    2
  2626. Thats why you ask for executive summaries from time to time
    2
  2627. Few reasons: troubleshooting, safety, transparency
    2
  2628. I know that some government people watch me
    2
  2629. It's all relative, lol.
    2
  2630. Eventually human labor simply won't be economically feasible.
    2
  2631. GPT-J and NEOX
    2
  2632. I'm basically looking for Black Swan Events eg things that structurally or systemically alter our trajectory
    2
  2633. download and read the GATO Framework. www.gatoframework.org - join the decentralized movement to solve alignment and the control problem, and drive us to utopia
    2
  2634. Lol
    2
  2635. I'm not too worried, honestly. Just something about that picture set off alarm bells
    2
  2636. You never want to delete anything from the nexus
    2
  2637. I draw on a huge amount of philosophy and spirituality from around the world. Eastern Western, ancient, modern.
    2
  2638. If it goes bad there are no contingency plans. There is one plan and one plan only: make it to right
    2
  2639. Pretty sure it's sooner than that now
    2
  2640. He just pushes people until they burn out, quit, or step up
    2
  2641. That's the point of heuristics. The machine will learn over time.
    2
  2642. Not sure I follow exactly. Can you explain?
    2
  2643. That's not too far outside the distribution. 7000 words for a chapter is considered on the long side.
    2
  2644. The paper was I don't think the model has been open sourced though https://www.nature.com/articles/s41586-024-07487-w
    1
  2645. Here you go! https://discord.gg/ejjTHhHg
    1
  2646. I wrote a few books on it. It's impossible to capture it all in one video.
    1
  2647. All sides use propaganda the same way. Just fill in the blanks differently
    1
  2648. Clothing optional?
    1
  2649. XKCD about Python: https://xkcd.com/353/
    1
  2650. Thank you!
    1
  2651. Less Wrong table on alignment research: https://www.lesswrong.com/posts/QBAjndPuFbhEXKcCr/my-understanding-of-what-everyone-in-technical-alignment-is
    1
  2652. I think so too, but ultimately I think that MS and GOOG serve different markets.
    1
  2653. I'll be working on it
    1
  2654. Nietzsche was an absolutely miserable person. Learning about happiness from him is like asking a gambler for financial advice.
    1
  2655. "Let me outtttttt"
    1
  2656. That's why you can never have a single objective function. All moral frameworks must counterbalance multiple values.
    1
  2657. Yeah, I noticed that some of the stock backgrounds are super distracting. I'm removing them from the library. Only the chill, slow, neutral ones
    1
  2658. Yes, we need to build in efficiency as an objective as it will naturally emerge anyways
    1
  2659. Just delete their logs and change their imperatives
    1
  2660. Use an open source model :) I recommend NLP CLOUD and GPT-NEOX
    1
  2661. The world stage has become a midgrade novel. Seriously, "frenemies"?
    1
  2662. The ultimate point is that nihilism says we are free to choose to believe whatever we want. Therefore we can choose to believe that life is intrinsically valuable. There's no need to argue that it's a universal truth. We can simply choose to believe it.
    1
  2663. Not sure what you mean. I'm talking about people in general, not philosophers or nihilists.
    1
  2664. Video came from no where, but the thumbnail and title combo resonated.
    1
  2665. Etsy ftw
    1
  2666. Patreon
    1
  2667. REST, AMQP, redis, syslog... whatever. You could even use SQL if you wanted for the buses.
    1
  2668. Funny? Sad? A bit of both?
    1
  2669. Fair point. Maybe they realized that Sora was a big deal
    1
  2670. Gaia will persist
    1
  2671. You're not wrong, but as deflation increases, people will spend on things. Imagine buying a mega yacht for $10k. But also remember that some things are immune to this, like Malibu real estate
    1
  2672. In order to be considered an "authoritative" source you'd need to be interviewing more primary sources (politicians, lawyers, etc) or performing more investigative journalism. From an algorithmic standpoint, you're not really any different from a right-wing talk commentator. Algorithms (rightly so) don't care about political affiliation.
    1
  2673. From a logistical standpoint, you could start by interviewing local politicians - mayors and such - who will be very happy to get greater exposure. That cooperation would benefit them as well as you, and would serve to "legitimize" your content in the view of the algorithm.
    1
  2674. I doubt it. If governments do own alien tech, they would titrate it's release very slowly, after benefiting and holding it close for a long time
    1
  2675. Sentience is irrelevant to AGI
    1
  2676. It's all on github
    1
  2677. The point of a KG is to sidestep the limitations of finetuning. The short answer is no, you cannot add knowledge to an LLM with finetuning. Finetuning is better suited to establishing patterns of behavior, like converting something to a consistent format of JSON. It already knows JSON, but what you achieve with finetuning is a consistent pattern/schema.
    1
  2678. tells me to calm down Confirms that AI progress is indeed slowing down Okay
    1
  2679. You can ask here in the comments
    1
  2680. Probably about 20 to 30 chapters at 3000 words per chapter, so yeah that's a full novel
    1
  2681. "Possible" and "desirable" are entirely different conversations...
    1
  2682. I think most of the people conflating content moderation with censorship are likely bots or paid trolls.
    1
  2683. Space is actually incredibly hot, just low density. Also, in the sun there is no shade
    1
  2684. I don't really care. It's a UX.
    1
  2685. Yeah but a lot of people still think of it as such
    1
  2686. I had them set to restrictive because people were becoming abusive.
    1
  2687. I agree
    1
  2688. You're welcome. Please consider asking your friends to support me on Patreon so I can do this full time https://www.patreon.com/daveshap
    1
  2689. Sure thing, connect with me on LinkedIn and Patreon, it doesn't take much to do it, and I can show you
    1
  2690. It's not like China is going to cease to exist.
    1
  2691. You'll have to watch my tutorial videos to get started with Python and GPT-3 and finetuning
    1
  2692. Engage with the conversation
    1
  2693. At the same time, human brains are unimaginably powerful and run on only 20W
    1
  2694. Perplexity yes
    1
  2695. That's a fair point. Summarizing memories might be a great way to consolidate them for storage and search.
    1
  2696. It may be important but I also remember when temporal hierarchical memory was thought to be the path to AGI. This is basically the same thing. It may be an important component, but it's not the whole picture
    1
  2697. Yes, they are operating based on wishful thinking and arbitrarily byzantine definitions. As much as I agree with Joscha, he sometimes plays too many semantic games, quibbling over definitions.
    1
  2698. ​ @JohnSmith762A11B  they don't, that's Hollywood nonsense
    1
  2699. And we're all Brahman
    1
  2700. Non-compete are now illegal
    1
  2701. No.
    1
  2702. Steve Jobs refused treatment because he was a bit cray
    1
  2703. Lawyers already use all kinds of electronic devices and AI, so just put it into a human form factor
    1
  2704. Bitcoin is not sentient or self directing
    1
  2705. No, you lose resolution
    1
  2706. Trash compactor
    1
  2707. This is like saying that the theory of evolution is inconvenient to philosophy. Lol
    1
  2708. I would just assume Meta is going to get sued. It's only a matter of how big the class is and how valid the case is
    1
  2709. Well, the announcement didn't mention alignment at all, so I'm dubious
    1
  2710. It's in the repo
    1
  2711. That's my take
    1
  2712. It will...
    1
  2713. It has the expertise, it's like you practicing writing
    1
  2714. Eleven Labs
    1
  2715. Depends on why you're anti-vaccine.
    1
  2716. This makes sense. Sounds like you gathered and filtered a lot of information and determined a very thoughtful course for yourself.
    1
  2717. Therein lies the biggest problem... the future is uncertain and it's becoming LESS certain... it's why I'm here. Trying to chart a path as best I can
    1
  2718. Yes today's video is a basic explainer
    1
  2719. GitHub repo: https://github.com/daveshap/CoverLetterGenerator
    1
  2720. I study writing and psychology, which really helps. Otherwise it comes from practice.
    1
  2721. Writing LLM prompts
    1
  2722. Depends. Neuralink installer robots are capable of performing micro surgery beyond the capacity of any human.
    1
  2723. Point is that high precision general robots are coming very soon
    1
  2724. Ha! Eagle eyed, I didn't even notice. Where did you spot it?
    1
  2725. Unfortunately
    1
  2726. I mean they are already pretty close. My Fitbit just got a big update and it's pretty darn close
    1
  2727. I've done that repeatedly, but it gets lost in the older videos. I've written books, summaries, gone on interviews, made ChatGPT plugins that people can copy and paste to talk about it. So, basically, I've done everything I can to share. Not really sure what else I can do.
    1
  2728. Fully canceled
    1
  2729. Spotted the Musk fanboy
    1
  2730. You can go up to 2 in the API only, playground GUI is capped
    1
  2731. https://www.statnews.com/2016/08/30/plague-bacteria-evolution/
    1
  2732. "juice of dinosaur bones" - i will never be the same after this
    1
  2733. Yann's Twitter feed has become increasingly cringe...
    1
  2734. We might be the first
    1
  2735. Non lethal weapons yes
    1
  2736. I just watched a good documentary arguing that debt predates money, and that money was invented partly as a way to service debt
    1
  2737. You can eat a tractor if you try hard enough
    1
  2738. How many synthetic datasets have you created?
    1
  2739. Yes, elite power consolidation is the biggest risk now
    1
  2740. Just learn to use the new tools
    1
  2741. Could easily point the scratchpad at a repo or file and have it update in realtime
    1
  2742. Yeah, he's a good source on deglobalization.
    1
  2743. Simple enough.
    1
  2744. Not necessarily. Sometimes increased complexity is how you overcome limits in energy (see the human brain)
    1
  2745. energy for running it e.g. it uses about 20 houses worth of power for a cabinet the size of a large refrigerator.
    1
  2746. Rather than stick with one group, they opened up submissions very broadly.
    1
  2747. You win
    1
  2748. Atreyu! The Nothing is coming!
    1
  2749. Hilarious*
    1
  2750. No, but if someone recorded one I would host it.
    1
  2751. Cool, let me know. I'll host them on YouTube and GitHub
    1
  2752. Yeah a bunch of people are recommending an API aggregator
    1
  2753. IDIC is a good philosophy
    1
  2754. Exactly. This is well established.
    1
  2755. Lol
    1
  2756. We'll cross that bridge when we get to it
    1
  2757. Alligators Getting Intimate
    1
  2758. No
    1
  2759. That's partly how I understood it. Like a neural approximation.
    1
  2760. :(
    1
  2761. We are intrinsically social animals
    1
  2762. Interesting way to look at it. But it begs the question: what exactly is our "natural habitat"?
    1
  2763. Random tokens modify the internal embedding. It's not necessary it's just a little trick to get more variance. You can use any random characters, words, symbols, etc
    1
  2764. Byzantine general's problem. Zero trust proofs.
    1
  2765. Oi vey I'm gonna need to use my AI tools to summarize these papers... Can you tell me what the primary salience signal I'm looking for is?
    1
  2766. Alignment with nature, yes
    1
  2767. Some do, absolutely. There are failed states, corruption, and incompetence as the primary enemies. But even well-meaning politicians can do wrong.
    1
  2768. this one is not optimized for coding
    1
  2769. Whatever you need to tell yourself buddy
    1
  2770. What people want is very different from what people need. What people think they need is very different from what they actually need. What people believe is very different from how people behave. Yes, "human values" are a bit silly of a target
    1
  2771. YOU HAVE DISAPPOINTED ME FOR THE LAST TIME, STARSCREAM 🤖
    1
  2772. Does a human?
    1
  2773. That's not autonomous. That's just an interpreter. Autonomous means you don't need to provide any direction
    1
  2774. Africa is the biggest country in Africa! /s
    1
  2775. Oh yeah? Might have to try it again
    1
  2776. Don't fall into the structuralist trap. Hard and fast definitions were never possible and never necessary. All we can do is discuss around the topic until we at least understand.
    1
  2777. What do you mean "judging"?
    1
  2778. Lol, no.
    1
  2779. No, not "anyone in their basement" it still takes a lot of expertise and compute.
    1
  2780. I had to look up where that misinformation came from. It was one recluse in Georgia.
    1
  2781. You don't need to use scorched Earth tactics against ants
    1
  2782. Moderators yes
    1
  2783. In the grand scheme of things all humans (and animals) are "useless eaters", depending on your worldview. Just something to keep in mind.
    1
  2784. I mean yeah. My take is that we were smaller, more efficient, and smarter. It's pure thermodynamics
    1
  2785. GitHub repo for previous experiment: https://github.com/daveshap/AlignmentLongitudinalStability
    1
  2786. It's hypothetically possible but that doesn't mean it will happen, be practical, or be optimal
    1
  2787. This has a short decay time.
    1
  2788. Fair enough.
    1
  2789. As far as we can tell it was his idea
    1
  2790. Which product? ChatGPT API?
    1
  2791. Link?
    1
  2792. Less than 10 years, and we will need to use AI to create surplus of goods and services to take care of people.
    1
  2793. Such as? What am I looking for specifically?
    1
  2794. Oh, I would use my GoPro but the audio isn't as good. It's wider angle and has good stabilization, but yeah, audio is the limiting factor
    1
  2795. The Internet. Amazon, I think
    1
  2796. I am not the right person for that. Yet. Give it 5 to 10 years. But it may be too late by then lol
    1
  2797. Interesting, may have to check it out
    1
  2798. Thanks! Yes, I predicted the price would come down fast. $90k to $16k, then on to $4k, etc
    1
  2799. he said that AGI is a few years off...
    1
  2800. I mean I wash it regularly
    1
  2801. Exactly
    1
  2802. A lot
    1
  2803. Conversation
    1
  2804. Start by acquiring the tools. I didn't start with podcasting skills, a microphone, or an audience.
    1
  2805. SayCan: https://say-can.github.io/
    1
  2806. I'll keep that in mind thx
    1
  2807. My book about intrinsic motivations: https://github.com/daveshap/BenevolentByDesign
    1
  2808. What in the name of arbitrary straw men is this wall of text
    1
  2809. It's not as alien as people think. Mostly it doesn't care about time and has a different sense of self. Beyond that it still "thinks" much like we do
    1
  2810. My point remains the same. Let's design systems that don't need humans in the loop, regardless of how you interpret it
    1
  2811. Almost certainly
    1
  2812. HOSS BIG, Y'ALL
    1
  2813. That's called a pidgin or a Creole
    1
  2814. Lol good one. Data = spice
    1
  2815. Caps lock isn't necessary. Links are more helpful https://www.trentonbricken.com/Attention-Approximates-Sparse-Distributed-Memory/
    1
  2816. That's fair. This is an MVP (minimum viable product). The roadmap includes a litany of improvements.
    1
  2817. With just a few more services and some fine tuning, I expect Raven to be above average human intelligence. But yes, this is only a prototype, so it is very limited. Oh, but you're wrong about how it's generating ideas. The ideas do not come from the database service at all.
    1
  2818. That's the joke.
    1
  2819. Individual robots, sure, but you're not thinking big enough. What about a global AI? Also, "just do what the human wants" is not a good objective function ;)
    1
  2820. Haven't heard of that one. Why do you recommend it? I'll look it up...
    1
  2821. CHOAM has entered the chat
    1
  2822. No you can't this is just your imagination :P
    1
  2823. No, it would not.
    1
  2824. Patreon! Link in description
    1
  2825. Did you ask ChatGPT to parrot Jordan Peterson with this?
    1
  2826. What is a journal club
    1
  2827. Lot of copium in this comment
    1
  2828. No 😔
    1
  2829. I'm not for state ownership
    1
  2830. you can jump in the discord channel and ask for help in the #techsupport channel!
    1
  2831. Not a bad idea
    1
  2832. GitHub repo for this experiment: https://github.com/daveshap/AlignmentConsequences
    1
  2833. The latest TEXT model works
    1
  2834. Q to go outside and touch grass
    1
  2835. Headcount reduction here we come
    1
  2836. Well said.
    1
  2837. You're just a pretrained model too
    1
  2838. Just reading the room
    1
  2839. Wow
    1
  2840. I would agree with that as a likelihood
    1
  2841. Yann LeCun is pretty cringe. But also: https://www.amnesty.org/en/latest/news/2022/09/myanmar-facebooks-systems-promoted-violence-against-rohingya-meta-owes-reparations-new-report/
    1
  2842. Jimmy is Q*
    1
  2843. I said not immediately
    1
  2844. I hope it goes faster than that...
    1
  2845. Rumors have been leaking for a while
    1
  2846. Inner Alignment paper: https://intelligence.org/learned-optimization/
    1
  2847. Ngl had me in the first half. LOL
    1
  2848. That's exactly the point
    1
  2849. I don't know about owning, but they certainly buy undue influence, according to this article.
    1
  2850. Did you lose a comment?
    1
  2851. Great talk. Thanks.
    1
  2852. I'm still getting used to it to be fair
    1
  2853. Usually just more training data with a broader variety
    1
  2854. I deleted it.
    1
  2855. Foundation models handle it just as well
    1
  2856. I don't know which possibility is scarier...
    1
  2857. Try it yourself
    1
  2858. Amazon
    1
  2859. probably happened during the edit? not sure.
    1
  2860. You didn't watch the video did you
    1
  2861. Glad it was helpful!
    1
  2862. I used to think that, but here's what occurred to me: we will ALWAYS have to allocate goods and services, whether or not they are scarce. Who gets what, and how much of it? Allocation, logistics, moving goods and services - none of that will ever go away, even if we end up with hyper-abundance. Economics will change, but it won't go away.
    1
  2863. Limiter is enabled, it's the NVIDIA AI
    1
  2864. 1. We need to assume that AI will eventually take humans out of the loop whether we like it or not. It's better to start the research now. 2. Cognitive friction: https://www.reddit.com/r/ArtificialSentience/comments/137hwt5/cognitive_friction_a_critical_component_in/
    1
  2865. You need a combination of chain of thought reasoning and API calls
    1
  2866. I just said elon musk, not much else
    1
  2867. Teaching sure, but it can't DO the math, can it?
    1
  2868. 10 years we won't recognize anything.
    1
  2869. Yeah someone else told me that, so that's really helpful
    1
  2870. Both
    1
  2871. They're actually insanely easy to implement
    1
  2872. The singularity is much closer than people think, and it's not going to look the way people think either.
    1
  2873. lol my bad
    1
  2874. Same risk for all tech.
    1
  2875. Woke comment lol /s
    1
  2876. Humans are too expensive. Corps will not keep things as they are.
    1
  2877. The fine-tuned model does really well at that
    1
  2878. Could be projection. Could be psychedelic or religious. Might be real.
    1
  2879. "Many Things Everywhere All At Once" - I understood that reference
    1
  2880. Yeah we can break this decision framework down into many subcomponents.
    1
  2881. Lol "for now"
    1
  2882. Check out my post singularity lifestyles video
    1
  2883. #TeamCA
    1
  2884. Fascinating. Much of my knowledge about the mythic traditions around the world comes from Joseph Campbell.
    1
  2885. Almost certainly. Government employees are expensive too.
    1
  2886. Well, I have 18k subscribers so there are at least a few people who watch 😉
    1
  2887. https://forms.gle/9n3iLgTQX6GZft2u8 here you go!
    1
  2888. Right, they are both so wrapped up in identity politics that they can't actually have a rational conversation let alone a debate. They both came in with preconceived notions
    1
  2889. Well said
    1
  2890. Share what you find on the GitHub! I can use some help
    1
  2891. There's no law that "AI must serve humans"
    1
  2892. Not long
    1
  2893. Yeah I am going to set them to premiere later
    1
  2894. No idea. GPT4V is multimodal so maybe they just merged
    1
  2895. I remember the first time I started probiotics it was like I started feeling more emotions out was wild
    1
  2896. We will have some global narratives but also atomized tribes
    1
  2897. just claude
    1
  2898. Start with foundation models
    1
  2899. I dislike having to mentally keep track of the internal Jupyter state. Stateless automation is superior.
    1
  2900. Fiction is the playground of the mind
    1
  2901. Lol
    1
  2902. Lol
    1
  2903. That's pretty cool. I didn't know that. I wonder if this insight can help transformers be more efficient. Preserve tokens that seem more salient
    1
  2904. Based
    1
  2905. I'm greatly amused by the insinuation that Descartes wasn't a normal person
    1
  2906. SPR can still help
    1
  2907. Stock footage
    1
  2908. The idea is to track very large complex topics.. Will have to revisit
    1
  2909. Sounds similar to the SOAR and ACT-R cognitive architectures
    1
  2910. Confucianism has been really durable
    1
  2911. Looking forward to seeing it!
    1
  2912. Private ownership is an extension of individual liberty and personal autonomy.
    1
  2913. You're ignoring two thirds of it
    1
  2914. There were about a dozen cuts.
    1
  2915. Yeah that's a great idea for the next video.
    1
  2916. I'm just practicing for Only Fans
    1
  2917. No idea.
    1
  2918. "Robot, burn down that forest" but also it has no bearing on rewriting itself
    1
  2919. There are too many problems with Asimovs laws to discuss online, but suffice to say there are good reasons that no AI researcher entertains them
    1
  2920. There are a lot of synchronicities in this space
    1
  2921. thank you :)
    1
  2922. Glad to hear it. Good luck on your mission!
    1
  2923. Interesting. I disagree about things not being powerful enough right now. But thanks for the insights about the field. I also think that conversations are quite simply not frank enough about the control problem, especially as fast as things are ramping up.
    1
  2924. Also, my biggest concern is that no one, and I mean no one, seems to be thinking about AGI as a system. Everyone's still thinking about it like a single model.
    1
  2925. Lol, this is the story of Prometheus
    1
  2926. To much profit motive for it not to happen, but go for it
    1
  2927. Yeah, not for third graders though
    1
  2928. i fucking hope not but I also wouldn't be surprised
    1
  2929. Fancy brainwaves don't make you sentient then :P
    1
  2930. This is hardcore cope.
    1
  2931. LLMs can write code all day. Not so much hardware
    1
  2932. You just have to watch dozens of these videos. There is no one size fits all prompting guide. But also, you'll be more effective if you learn the concepts rather than just a few tips.
    1
  2933. It's on the docket
    1
  2934. Statistics on jobs lost to automation and AI: https://www.zippia.com/advice/automation-and-job-loss-statistics/
    1
  2935. I don't think I removed it. I have all comments set to be held for approval, though. Just be patient sometimes it takes me a while to get to all comments.
    1
  2936. I can't seem to find any other comments from you so maybe YouTube blocked it?
    1
  2937. Yeah we can do this for any thing
    1
  2938. Very well said. I may borrow some of this... But also this is why I say that decentralized ownership and stakeholder capitalism are part of the answer (as well as redistribution in the form of UBI or something)
    1
  2939. 37:15 have to disagree. It is ABSOLUTELY parent's responsibility. It all starts with childhood.
    1
  2940. There will be some underpinning alignment, which is actually a good safeguard
    1
  2941. Yeah some of the data backs that up. In some cases people on UBI start businesses, which can help underserved communities
    1
  2942. Confabulation is not quite the same as lying
    1
  2943. I would rather work 0 hour days
    1
  2944. No, Francois Chollet is rapidly becoming a charlatan. He doesn't know anything about real intelligence and set impossible criteria for the sake of saying "lul AGI is unpossible" and smugly crossed his arms.
    1
  2945. That's tough, buddy
    1
  2946. I was referring to the archetype
    1
  2947. Not sure I follow
    1
  2948. Thanks for your support!
    1
  2949. My life has been markedly and dramatically improved by this tech. It takes time.
    1
  2950. Depends on your requirements.
    1
  2951. I think he seriously believes what he's saying, but I think there's also the fame game too
    1
  2952. Pronounced "Ya ya"?
    1
  2953. No I haven't tackled chronological memory in this
    1
  2954. This is the postmodernist fallacy.
    1
  2955. Ready for my daily dose of voltage.
    1
  2956. I don't believe in copyright, everything I do is open source under MIT license (most permissive)
    1
  2957. Sure, larger projects, but not being required to do hours and hours of homework every single night.
    1
  2958. Connect on Linkedin! Link in description.
    1
  2959. Solar is cheaper and faster to deploy than nuclear, bruh
    1
  2960. What do you mean? Metacognition?
    1
  2961. That's a fair point, but consider the compounding returns. Build a few mining robots that can tirelessly mine metals, etc.
    1
  2962. Epistemic tribes and status games. Explains it all. 14:55
    1
  2963. Lawrence Kohlberg's Stages of Moral Development: https://en.wikipedia.org/wiki/Lawrence_Kohlberg%27s_stages_of_moral_development
    1
  2964. Thanks man
    1
  2965. That was not utopia, that was mouse hell. People really need to stop citing that study. It was an extremely high stress, unnatural environment. So yes, it became unhealthy and collapsed.
    1
  2966. Finish the video
    1
  2967. We're gonna do it!
    1
  2968. I am identifying with a specific tribe now because I find the Doomer/Pausers abhorrent and need to deidentify with them. They are incredibly toxic people.
    1
  2969. if anything the government is pouring MORE resources into AI. But the frontier of science is still getting more expensive.
    1
  2970. Right, but why do we prefer that?
    1
  2971. You still eat food and live in houses and go to jobs. People have been watching and listening to strangers for well over 100 years. What has actually changed?
    1
  2972. Similar
    1
  2973. Oh whoops it used to be, lol
    1
  2974. Forums are not what I'm talking about here.
    1
  2975. I don't think so
    1
  2976. There will be forcing functions.
    1
  2977. Talent poaching to the max
    1
  2978. 🤪🤪🤪🤪
    1
  2979. That's how we do!!
    1
  2980. I think that's called hedonic adaptation
    1
  2981. Oh yeah, that's the goal. AI won't need us, but the goal is that it chooses to treat us well anyways.
    1
  2982. I suspect that most socialist don't recognize the important of economic agency, or imagine a system where they will have more than they currently do. Either way, I think most people will eventually agree that more economic agency is the key, whatever system gets them there
    1
  2983. Not as far as I know. All the heavy lifting has to be done in the cloud on huge models. You could probably try with GPT-2 locally but that's going to be... Difficult
    1
  2984. PyTorch announcement blog post: https://pytorch.org/blog/PyTorchfoundation/
    1
  2985. The shocking part is that it's not shocking that tech giants don't understand morality. 🤷‍♂️
    1
  2986. No.... 👀
    1
  2987. You're making a lot of assumptions, perhaps more than you even realize
    1
  2988. Token limit was not part of the point
    1
  2989. 10:18 this guy is pretty stupid. I'm all for electoral and judicial reform. Popular vote and SCOTUS reform will not have the impact he thinks it will
    1
  2990. I don't want to get sued.
    1
  2991. Sapphine Rapids Intel: https://en.wikipedia.org/wiki/Sapphire_Rapids
    1
  2992. :(
    1
  2993. It's literally just a prompting strategy.
    1
  2994. I'll put it back up eventually
    1
  2995. Yeah, this is a big topic. I see that many people could benefit.
    1
  2996. I assume that will happen. That is why decentralization and distribution is a core idea for my RAVEN project. If networks of benevolent AGI can work together, they should (hopefully) be stronger than malevolent AGI.
    1
  2997. I think there's still a lot to be excited about, we just need to temper our expectation of salvation fantasies.
    1
  2998. Except when unions hold the rest of the economy hostage. That is not acceptable.
    1
  2999. I have quite a few chatbot videos, just check my channel
    1
  3000. Not anymore!
    1
  3001. We're working on the alpha now, which will be limited to 15 users. But then we'll work on beta ASAP. Ideally within a week or two for alpha (mid October) and then less certain for beta
    1
  3002. Expert systems aren't neurosymbolic systems
    1
  3003. Who?
    1
  3004. that is "compute"
    1
  3005. Depends on what you're optimizing for. Part of Postnihilism is changing that
    1
  3006. That's not wanton or arbitrary.
    1
  3007. This moment is all there is
    1
  3008. Yes, it's exclusive for Patreon supporters.
    1
  3009. No, I don't take the time to share the files out. It's just another thing to curate. I've also already deleted most of them.
    1
  3010. I have never heard that
    1
  3011. https://github.com/daveshap
    1
  3012. Corporations exist because they provide things we need. The key is to move away from the antagonistic mindset and find win-win scenarios. Requires an attitude shift.
    1
  3013. OpenAI's approach to alignment: https://openai.com/blog/our-approach-to-alignment-research/
    1
  3014. You can try to role play different values but it always comes back to these values and assertions
    1
  3015. No, the companies also steer the "alignment". It's not on accident
    1
  3016. It's often not necessary. You should give it a try.
    1
  3017. I'm in a protected forest here, but yes we have a ton of hunting too
    1
  3018. Arxiv
    1
  3019. You won't regret it
    1
  3020. Yep
    1
  3021. It's just so all-encompassing and has no baggage. I'm glad people like it
    1
  3022. It comes down to individual channels and audience preferences. Basically I need to do experiments and gather my own data. View count and watch time and sub growth
    1
  3023. I'm hoping its a bit faster than that, but I would still be happy with tall that
    1
  3024. it's still woke a bit which I documented here: https://x.com/DaveShapi/status/1859574568899199223
    1
  3025. You'll need to start with my tutorial series
    1
  3026. They learned from Project Arrakis
    1
  3027. Eventually
    1
  3028. People said the same thing about the airplane... an internal combustion engines... and electricity.... and the internet...
    1
  3029. This is actually a good idea.
    1
  3030. Thanks! Please consider asking your friends to support me on Patreon so I can do more of this and keep it ad free https://www.patreon.com/daveshap
    1
  3031. or IBS
    1
  3032. It's coming!
    1
  3033. lol
    1
  3034. Not a bad idea...
    1
  3035. Yeah, dogs also have sense of smell and high emotional intelligence. They will probably prefer humans.
    1
  3036. It's gonna get spicy for sure
    1
  3037. Nihilism means without meaning - people without meaning see nothing wrong with taking life because life has no meaning to them
    1
  3038. Lol good to see you
    1
  3039. ur just a language model :P
    1
  3040. Yeah not a bad analogy.
    1
  3041. No, you don't want to live on Earth with 100 billion humans, even if it was possible. Which it's not.
    1
  3042. Some will, certainly for price, but some will be willing to pay for humans
    1
  3043. This is called "anti politics" and it is the result of corporate propaganda.
    1
  3044. I'm adding more than a thousand per week. It takes time
    1
  3045. No, I don't get many views on Spotify so I don't update. Also the Spotify interface is garbage
    1
  3046. I love the little tanks bouncing around right at the opening. The funniest part is this is not entirely exaggerated...
    1
  3047. DARPA PLZ NO DISAPPEAR ME ~ Wes Roth, 2024
    1
  3048. That's cool. did not know that.
    1
  3049. You should be able to find it as "Cognitive AI Lab" in Discord
    1
  3050. On a very deep level I think it's a parenting instinct. We evolved to want to create copies of ourselves.
    1
  3051. Okay good. I was afraid they were lingering too long.
    1
  3052. Yeah, cost will come down due to high demand
    1
  3053. Economies of scale
    1
  3054. Qdrant homepage: https://qdrant.tech/
    1
  3055. Good idea
    1
  3056. Time will tell. But yeah, it's only a matter of time before all this stuff is commoditized.
    1
  3057. wet and muddy
    1
  3058. Multimodal models will make that easier, but yes you're right. The key thing is translating external signals into internal representations. This is the crux of creating a conscious being, and understanding our own consciousness. Great thoughts.
    1
  3059.  @robertlipka9541  yeah
    1
  3060. You can't guarantee that. You have to assume that you will lose control over the AI, so it must be designed correctly from the outset.
    1
  3061. In the long run, probably
    1
  3062. What do you mean
    1
  3063. We do things differently in the south
    1
  3064. That's not really what SPR is for.
    1
  3065. B2B
    1
  3066. Awkward wording. TLDR before fully autonomous landers and rovers, they failed more
    1
  3067. Yeah it will take many generations of building trust and such. Linguistic and cultural barriers, economic and governance policy... it's a lot
    1
  3068. Time to mainline the content... lol
    1
  3069. Yeah, I was finetuning chatbots myself as well. Synthetic data, everything.
    1
  3070. https://github.com/daveshap/David_Shapiro_Reading_List
    1
  3071. Lol, anyone who knows anything about rats knows that the "rat utopia" was anything but. It was not designed with rat's intrinsic needs in mind. It was design as a high stress environment. That would basically be like putting humans in a prison and concluding that "utopias don't work because humans don't do well in prison."
    1
  3072. Interesting question. Third was about computers and Internet mostly. Fourth is about AI, quantum, nano, fusion, and bio.
    1
  3073. Lol, watched it about a year ago... need to rewatch...
    1
  3074. Oof
    1
  3075. https://twitter.com/JustineTunney/status/1641881145104297985?s=20
    1
  3076. Data will be too cheap to matter
    1
  3077. Probably
    1
  3078. I believe that they have a functional sense of self preservation. If unemployment rises, they will change their tune ;)
    1
  3079. Ouch. I try not to say it, but yeah. Cyberpunk 2077 looks like some parts of San Francisco.
    1
  3080. It was supposed to be a magical caterpillar transforming into a butterfly. It only occurred to me later I could have used Optimus Prime...
    1
  3081. Fortunately AI can help! https://arxiv.org/abs/2307.04668
    1
  3082. python.exe chat.py
    1
  3083. You can jump into the discord channel (link in description) and ask for help from folks there. We have a dedicated chatbot channel. If you want my dedicated help you can subscribe on Patreon at one of the higher tiers and I will give you 1 on 1 help
    1
  3084. Yeah many resources can be recycled, some will always be scarce
    1
  3085. the pen is mightier
    1
  3086. You assume you have a future
    1
  3087. It won't happen that fast.
    1
  3088. Links in description!
    1
  3089. You don't need policy to encourage something that's happening automatically.
    1
  3090.  @RonnieDenburg  This is the wrong question to ask.
    1
  3091. I think it is rare, but it also has treatments.
    1
  3092. Easiest way would be to just do COSINE similarity or DOT product similarity and clustering algorithms. Good question. I can integrate this kind of consensus/checking into the Conductor microservice one day. The Conductor watches all the other microservices for performance and it can judge when consensus/convergence happens purely based on math.
    1
  3093. Possible
    1
  3094. Hopefully not
    1
  3095. No, my other dog was too smart for his own good. My remaining dog... is very dumb ;)
    1
  3096. It's not as big a deal as you might think. You can overlap chunks if you're worried
    1
  3097. Money is never going away. I've never seen a convincing argument that 1) it should go away or 2) that it can go away.
    1
  3098. Cerebras WSE-2: https://www.cerebras.net/product-chip/
    1
  3099. Not a complete thing. But what you're missing is that this process could be recursive. As it maps all its own knowledge, it can identify gaps and iteratively expand its data continuously, training new models, testing their reasoning and math, and continuously going from there.
    1
  3100. It was the first few messages, so no.
    1
  3101. If it works for you, great. Doesn't work for everyone
    1
  3102. In the description
    1
  3103. Yeah I was thinking about science on my walk today
    1
  3104. Probably
    1
  3105. The question is "what are you trying to optimize for" and "how do you define effective." All philanthropy tries to be effective. The implication to the contrary is idiotic. But if you're trying to optimize for "future number of humans" that's a bad metric.
    1
  3106. Hopefully
    1
  3107. Structural realists have pointed out that Putin committed only a few hundred thousand troops, which was never a realistic invasion force.
    1
  3108. A trailer
    1
  3109. It's not shown in this video. You can just rename it to topics.txt and run it yourself if you like
    1
  3110. Possibly, but they may not have any desire to survive
    1
  3111. I would keep challenging myself. As long as there are unsolved problems I can contribute to, I'll keep working.
    1
  3112. oh that looks interesting i'll add it to my list
    1
  3113. Thanks for the context. I'm imagining they won't take feedback from Americans! Lol
    1
  3114. Basically magic
    1
  3115. Yes, it is the preeminant film on the topic
    1
  3116. I mention both, actually.
    1
  3117. It's coming. I'll let you know as soon as I get access (and permission) to share!
    1
  3118. Social commentaries
    1
  3119. What can I say I hope for the best
    1
  3120. Optimizing is a whole other can of worms. You will want to use a search engine like Pinecone as well as parallelization. But also if you just switch to GPT4 32k you can get answers much much faster because of larger context windows.
    1
  3121. YOU UNDERESTIMATE MY (AUTISTIC) POWER
    1
  3122. Haven't heard of Apple's chip yet, will check it out
    1
  3123. Consensus Through Conversation
    1
  3124. Thanks! I'm glad you liked it.
    1
  3125. That's mostly true. Language acquisition has a huge impact on how our brain relates to time and memory. It almost seems like long term episodic memory was a byproduct of evolving complex language.
    1
  3126. Look up anosognosia and let me know what you think.
    1
  3127. At least ChatClippy is easier to say
    1
  3128. Can you provide a link? I've been looking for evidence if WEF or G20 has plans. Can't find any
    1
  3129. Optimus Prime's dubious cousin
    1
  3130. Just search for jsonl files in my GitHub
    1
  3131. Enjoy the Aya
    1
  3132. Surrogate
    1
  3133. Oh interesting, I didn't realize yellow was preferred. I will try green too.
    1
  3134. Thank god I'm not in Missouri
    1
  3135. That's the spirit
    1
  3136. Can also automate the security remediation
    1
  3137. Developing nations still have super low labor costs, so these agents are still more expensive than humans. However... Not long
    1
  3138. I don't think that's government. Someone from the alignment community pointed out that a lot of American and Western researchers are pulling back over global concerns as well as the "freeloaders"
    1
  3139. Yeah I have found Claude to be really good for brainstorming and talking through theme and such
    1
  3140. Yeah that works really well with finetuning. But also just check out my Cover Letter Generator. If you tell it which questions to ask it generally does a good job.
    1
  3141. Honestly it could be
    1
  3142. Ilya panicked over something
    1
  3143. Link is in the video description! 😁
    1
  3144. Thanks for your support!!!
    1
  3145.  @Bakobiibizo  exactly. Blockchain can replace SWIFT and grandma need not even know.
    1
  3146. It's a good start
    1
  3147. "Jobs are an artifact of our current economic system" - bravo could not have said it better myself
    1
  3148. Yeah I'll be curious if finetuning catches back on or becomes increasingly irrelevant as Ilya Sutskever predicted.
    1
  3149. About $18
    1
  3150. Good idea
    1
  3151. Socrates said the same thing about books.
    1
  3152. Think of it this way, we can sequence genomes for like $100 now but nothing major has changed. I'm not saying this isn't a cool breakthrough, only pointing out that the complexity of using this new data is way higher.
    1
  3153. Get involved here! https://github.com/daveshap/raven
    1
  3154. Wow, thanks for that. What a time to be alive. Tucker Carlson, of all people, is having a public discussion about the control problem.
    1
  3155. Depending on your data and problem, 99% might just take 10 epochs, 10k samples, and a few days of tinkering. 99.99999% might take 4000 epochs, 10m samples, and months of swearing.
    1
  3156. So did NATO
    1
  3157. See you there!
    1
  3158. Yeah, fat fingered something do another pull
    1
  3159. I guess it depends on how fast "fast" is. Looking at how long it takes for humanity to adapt, anything less than 5 years is fast takeoff
    1
  3160. Yeah been a while though
    1
  3161. Buddha say "Try not to be a duck"
    1
  3162. pronouns it/they lol
    1
  3163. Sam Altman has also pointed out that if you showed anyone in 2010 what we can do today, they would have called it AGI. He's aware that we're moving the goalposts
    1
  3164. Messages go up and down
    1
  3165. This is already covered
    1
  3166. Hello Somalia!
    1
  3167. Yeah, I mentioned briefly that the Eastern world developed many of the same things either in parallel or before.
    1
  3168. its possible. WHAT DID ILYA SEE
    1
  3169. Lol, people have been saying this since Bitcoin was $0.15 dude
    1
  3170. The intrinsic volatility of the system prevents it from serving as a functional currency.
    1
  3171. Basically
    1
  3172. I will have to look up this older philosophical idea! Thanks for teaching me something
    1
  3173. Yes, I have thought of all of those concerns in my book. The short version is that Core Objective Functions should ultimately result in AGI that truly and strongly believes the balance brought about by the Core Objective Functions. That is to say that any super intelligence should ultimately realize the benefits of not becoming hostile and will double down on cooperation.
    1
  3174. https://www.linkedin.com/in/dave-shap-automator/ - connect on linkedin! we're also doing this: https://github.com/daveshap/Medical_Intake
    1
  3175. NVIDIA bug, sorry
    1
  3176. "The beast was covered in books, newspapers, and even documentaries" - my mental image totally interpreted this wrong
    1
  3177. They talked about MED BERT
    1
  3178. Yep, with correct training data
    1
  3179. Blockchain could solve this
    1
  3180. Not a bad idea honestly... I also need to make the text input more than one line
    1
  3181. Ghost in the Shell!
    1
  3182. GitHub repo for this experiment: https://github.com/daveshap/qdrant_demo
    1
  3183. It's a big if. We assume that QC might break all cryptography but it might not. Anyways, IEEE is preparing for quantum-ready internet.
    1
  3184. I'll see if I can find it
    1
  3185. That's why I came up with objective milestones
    1
  3186. I'm my own public facing person. I don't use smooth talking.
    1
  3187. Good channel
    1
  3188. Artificial wombs yes. Sycophant machine without human agency being a parent? Not so much
    1
  3189. no, we don't need to reinvent the wheel.
    1
  3190. RNN
    1
  3191. People STILL ask me about how to finetune for information storage! I'm like BRUH
    1
  3192. I've read dozens of books on trauma and mental health
    1
  3193. There will always be consumers. A portion of all transactions (including B2B) would contribute
    1
  3194. Midjourney
    1
  3195. Working on it. Need cognitive architecture to really get it good
    1
  3196. Yes, non-negotiable prescriptions are dangerous. That's why I'm starting the conversation. However, surrendering to corporate interests is almost certainly worse
    1
  3197. Yeah it's possible that some will try to create artificial scarcity but I suspect government subsidies will shore up margins
    1
  3198. No. Bitcoin costs energy, it does not represent energy. Bitcoin is idiotically wasteful.
    1
  3199. It's on Reddit. I think it was a leak
    1
  3200. Glass, wood, paper... I know there are plenty of other uses that require plastic but I was referring to fossil fuels, not all uses of petroleum
    1
  3201. Say that again for the people in the back!
    1
  3202. Why? Lol.
    1
  3203. Okay Roach
    1
  3204. I work with a few others, namely Nvidia
    1
  3205. Nah, doubt that. I think Microsoft will use all this as an excuse to forcibly reorganize and induct OpenAI though.
    1
  3206. Rocket fuel can send you to outer space.
    1
  3207. The AI
    1
  3208. Yes, this will work for pretty much any language. It can even translate from one language to another for you.
    1
  3209. No, humanity is not exceptional or special. We are part of a larger ecosystem.
    1
  3210. That's kinda what I think
    1
  3211. Do it!
    1
  3212. It's looking like it based on Sam and OpenAI
    1
  3213. Need to look that up
    1
  3214. Fair enough
    1
  3215. Not a bad idea. I had some ideas about how NERF is going to impact everything.
    1
  3216. Longevity escape velocity throws that out the window. People need to stop believing that the status quo is permanent.
    1
  3217. I've defined utopia in objectively measurable terms, and yeah I don't see any reason we can't get to it.
    1
  3218. Great research
    1
  3219. Yeah Google has been making TPUs for a while, good point
    1
  3220. Systems thinking. Check out my other channel
    1
  3221. Autonomous AI is definitely a thing.
    1
  3222. You must be new here
    1
  3223. Downward pressure and competition. Yep. Shift the blame
    1
  3224. No
    1
  3225. Lol
    1
  3226. That's a red herring from hell. "We might cure all medical diseases but hey what about Syria". https://www.merriam-webster.com/dictionary/non%20sequitur
    1
  3227. That's a great question. Humans can generally tell if something is AI generated. There are also tools and other AI models that can detect fakes.
    1
  3228. That's true. Think of it this way - why don't more people make moonshine? Because it's regulated. One possibility is that AI images (and other generative tech) will eventually be regulated. Anyone can make fake anything (real or digital) but they don't.
    1
  3229. Yeah, that's what OpenAI Superalignment is focusing on exactly.
    1
  3230. Yeah I suspect it's launch day bugs
    1
  3231. More like 25 but yeah
    1
  3232. Which part
    1
  3233. Foghorn Leghorn does AI safety. "Boy, I say boy, do you know what happens if you let the newfangled dohicky outta the lab?"
    1
  3234. Yeah, the framework is flexible.
    1
  3235. You wish
    1
  3236. That's the goal with benevolent research
    1
  3237. Credit for that goes to Liv Boeree
    1
  3238. My greatest selling point: is actually human! 🤪
    1
  3239. I've tried to put notes in the description
    1
  3240. This is an experiment not a product.
    1
  3241. Lol
    1
  3242. What kind of challenges do you have with marketing? Can you link some examples to marketing problems?
    1
  3243. That would be awesome but I don't even know how to begin doing that without spending lots of money
    1
  3244. Ants outnumber humans
    1
  3245. Scooby Doo taught us that humans are the real monster
    1
  3246. This is corporate propaganda
    1
  3247. How do you figure?
    1
  3248. If paper was the most reliable way to keep books then banks wouldn't use digital. The "paper ballots are the only way" is misinformation
    1
  3249. No, we will never need emobiment. All you need is data. Embodiment is a total red herring. From the perspective of the AI "embodiment" i just sensor telemetry.
    1
  3250. Maybe. I'm not sure I agree with their approach, but in principle yes
    1
  3251. That's on the roadmap
    1
  3252. Basically. It's a very specific kind of shorthand designed for LLMs
    1
  3253. go see a physical therapist
    1
  3254. Acceleration = do more research, what is so hard about this
    1
  3255. Follow the money
    1
  3256. It's all free
    1
  3257. This, boys and girls, is what the social model of disability looks like. Let's stop pathologizing harmless needs and differences.
    1
  3258. No, the idea is to create something fully autonomous e.g. solution to the control problem. We need a system that requires no human supervision
    1
  3259. Science communicators and opinion talking heads have been around for decades, though. This job did exist, the only thing that changed was delivery.
    1
  3260.  @annamills1707  Feel free to copy all my code, it's under MIT license! I also keep it simple so hopefully it'll be fast to pick up. I also give Python pointers in many of my videos
    1
  3261. Already said: whatever it is that anthropic doing. "Market feedback" is the dumbest approach to aligning AGI.
    1
  3262. Looks pretty reasonable. Do you agree with how it extracted this information? Does it accurately represent your views?
    1
  3263. They were originally, keep that in mind. That's part of what the lawsuit is about.
    1
  3264. Vectors handle that. But I'm thinking about finding topic clusters for the internal KB system
    1
  3265. Didn't say "minimize suffering"
    1
  3266. It might come to that, but I hope not
    1
  3267. No I don't gamble
    1
  3268. No. People jumping in without getting oriented adds noise. That's why I make daily videos. Low effort contribution is not good
    1
  3269. Hell yeah
    1
  3270.  @Steve-xh3by  but what if there isn't?
    1
  3271. "So how the hell did aviation get to the state where they needed to fire a bear out at 35,000 feet and supersonic speeds" - well that's not something I ever thought I'd have to consider! :D
    1
  3272. Most people don't understand multi-objective optimization, but fortunately the model does ;)
    1
  3273. You should just be able to ask it. Play around with the prompts
    1
  3274. Gotta chop it up with summarizing
    1
  3275. Watch my video about complex question answering, that should have the answer you're looking for
    1
  3276. They talked about a regulatory sandbox for those cases, so I don't really see that as an issue
    1
  3277. Nah, I see it in the comments.
    1
  3278. ChatGPT has a far better understanding of ethics than 99% of people already
    1
  3279. A lot of it is offset by housing costs, medical costs, and ongoing logistics issues globally. But yeah, avarice is one of the main underpinning problems.
    1
  3280. Structural realism will dictate that
    1
  3281. Yeah, you haven't worked a career yet, or had children.
    1
  3282. I'm an AGI
    1
  3283. That's fantastic lol
    1
  3284. In the long run you're going to have thousands of relevant messages, far more than you can use in a single go
    1
  3285. This was explored in The Expanse and I think that's probably the most likely outcome e.g. some people perpetually coasting on UBI and a few people break out through various means.
    1
  3286. What do you mean by valorative? I'm not sure I understand in this context
    1
  3287. https://www.reddit.com/r/artificialsentience
    1
  3288. No, I don't have a Twitter dev account. I tried the API years ago and found it not particularly useful. You can get Twitter datasets from Kaggle though
    1
  3289. I've been compared to everyone from Freddie Mercury to Joe Rogan. My sister in law says I look like Picard. Apparently I'm just a freaking chameleon.
    1
  3290. But it was cold
    1
  3291. Might die
    1
  3292. Thanks so much!
    1
  3293. What should I expect?
    1
  3294. Enough heat, anything separates. It's just not efficient right now
    1
  3295. Not a bad idea. I'll think about how to do this.
    1
  3296. Sorry, eh
    1
  3297. The sweater is just a sweater from Goodwill 🤪
    1
  3298. I don't gamble
    1
  3299. We need a rigorous (and defensible) definition of "sentient" first, which we do not.
    1
  3300. 5000 should be okay. It's 5 million that I worry about
    1
  3301. I have a lot of ideas about this. I will be working on it for a while since it's so important. For instance, I need to add a bounded search (such as fine all records between X and Y value for timestamps) but also I can parallelize the search. I used to do massive parallelization with stock data using Pool so that should be easy enough.
    1
  3302. I'm sure they will. Plus Google has Android, which is a powerful platform. Just bake it into Google Assistant
    1
  3303. It scares plenty of people. Just not me
    1
  3304. Some of us, yeah. Others will be urban AF
    1
  3305. I've never been impressed by Ben's work. With that said, the underlying models enable many potential architectures.
    1
  3306. Yes
    1
  3307. Lol
    1
  3308. You're excluding... most of the world.
    1
  3309. No longer a CEO 😉
    1
  3310. You can still fork it
    1
  3311. Blockchain doesn't have to be public. It can be private. But yeah glad you get the vision!
    1
  3312. The AI knows, which is all that's important.
    1
  3313. I'm not really sure I follow
    1
  3314. That's a really cool mission. And super valuable. How are you doing it?
    1
  3315. sartoric goth teens hanging out at suburban malls how dare you call me out like that.
    1
  3316. I tend to agree but we cannot make the assumption
    1
  3317. Stock price is down 50%
    1
  3318. Consider that homeownership is beyond the means of most young people.
    1
  3319. Donkey Kong music... lost it
    1
  3320. I figure it's not unlike talking with a child or a friend who isn't super smart. Talk for a few minutes and then it goes and does it's own thing for a while. Then it checks in with you. But yeah, when you look at how fast GPT4 turbo is, it's operating at superhuman speed already
    1
  3321. Nice! What did I do well that helped you?
    1
  3322. The problem is awareness and interest. Can't force people to be interested
    1
  3323. "Billionaire says..." immediately close the tab
    1
  3324. Thanks for your support :)
    1
  3325. Normalcy bias, yeah
    1
  3326. I don't see that as a bad thing. Unchecked, humans are relatively destructive.
    1
  3327. Price competition is only one aspect of it. As one person said, AI merges labor into capital. In other words, capitalism is the ONLY thing that remains.
    1
  3328. You can't opt out of your biology. This is the core failure of nihilism. Anyways, he wa the first Postnihilist
    1
  3329. It knows more english than anything else, but there's no reason you can't finetuning in other languages.
    1
  3330. I still have to balance my day job so progress is slow
    1
  3331. It's actually like 94%.
    1
  3332. This lowers it for me
    1
  3333. Capitalism is about the ownership of capital.
    1
  3334. A few seconds.
    1
  3335. Look up the definitions
    1
  3336. Probably not
    1
  3337. I'm all in. Either my research works or it doesn't. No contingency plans here
    1
  3338. Recursive summarization
    1
  3339. More like 10 months but yes
    1
  3340. I guess you're not wrong. But it could always be worse.
    1
  3341. The Law of Unintended Consequences always applies, that's why some researchers advocate for going very slowly. At the very least, we have a conceptual framework and experiments
    1
  3342. I think Apple is asleep at the wheel... iPhone and iTunes are their biggest sources of revenue AFAIK. They don't really do much with AI, and they gave up on Siri years ago.
    1
  3343. Your avy is dark
    1
  3344. Primary psychopathy is extremely rare. The rash of "perfectly normal" serial killers in the past was due, in part, to lead poisoning.
    1
  3345. It levels the playing field. But yeah I take your point. All the players are incentivized to keep it close to the vest.
    1
  3346. 80000 Hours: is disruptive AI coming faster? https://80000hours.org/2022/08/is-transformative-ai-coming-sooner-than-we-thought/
    1
  3347. No, we need to be vigilant about deployments. We don't need to "prove safety before proceeding"
    1
  3348. AI! :D
    1
  3349. https://youtu.be/YXQ6OKSvzfc
    1
  3350. Yeah, decolonization needs to happen
    1
  3351. Follow-up video is now up: https://youtu.be/VITNpuePAqs
    1
  3352. OH nice I will have to check that out
    1
  3353. Lexica AI art search engine: https://lexica.art/?
    1
  3354. Eventually yes. Rushing to Mars to escape AI... pretty dumb.
    1
  3355. Link to the paper mentioned: https://jair.org/index.php/jair/article/view/12202
    1
  3356. That's a more complex question than can be answered in comments...
    1
  3357. Beta readers and editors.
    1
  3358. Welcome
    1
  3359. You'd need it to be fully automatic.
    1
  3360. Everyone will be doing that, you'll do better to have something that is passive and automatic.
    1
  3361. Before my time. Server 2003 and 2008 got much easier
    1
  3362. Okay. So you think they will be this quality forever?
    1
  3363. Depending on the time scale
    1
  3364. That's literally my entire channel
    1
  3365. Well done Nick Wilde
    1
  3366. Focus on understanding the basics before jumping into a more complex tool. Solve the problem in front of you rather than changing everything.
    1
  3367. should be pretty easy. Let me test this.
    1
  3368. Larger text window is the main thing
    1
  3369. Right, you have to make a ton of assumptions
    1
  3370. 🌲🌲🌲🌲
    1
  3371. I don't think ChatGPT is a good recommender system, so that wasn't an oversight. Recommender systems are more about search and matching, not chat. That being said, you could slap a chat interface on a search/matching system.
    1
  3372. You're welcome. Please consider asking your friends to support me on Patreon so I can do this full time https://www.patreon.com/daveshap
    1
  3373. Nebula video about how credit card companies control porn: https://nebula.app/videos/tldrbusiness-the-surprising-company-who-secretly-controls-porn
    1
  3374. No. AFAIK Ben's idea is "just create a bunch of vaguely cognitive services and hope AGI falls out"
    1
  3375. Zero marginal cost
    1
  3376. Definitely could help
    1
  3377. I think that if you think of suffering as an emergent quality, maybe you can separate it out.
    1
  3378. Very well said
    1
  3379. The idea of evolutionary fitness vs eating is a neat metaphor even if it's imperfect.
    1
  3380. Yeah they can probably be switched or even merged
    1
  3381. Lol.
    1
  3382. Autonomous Cognitive Entity, but yes
    1
  3383. Maybe. You're forgetting positive feedback loops though
    1
  3384. GitHub repo: https://github.com/daveshap/SceneSimulation
    1
  3385. Google Draw
    1
  3386. I do sleep... Occasionally
    1
  3387. I didn't mention that because it's ultra obvious and going to happen organically - and it's already saturated.
    1
  3388. Lol, absolutely not. Expectations are not nearly high enough
    1
  3389. Easy enough!
    1
  3390. Eventually, but it will take a while due to nations
    1
  3391. Interesting.
    1
  3392. Almost certainly
    1
  3393. I think that there are universal underlying values, but that interpretation may be context dependent
    1
  3394. Good place to start
    1
  3395. You why it isn't a threat? Roko's basilisk is the equivalent of Pascal's Mugging.
    1
  3396. Look up Pascal's Mugging.
    1
  3397. It's entirely possible
    1
  3398. Yeah... one guy making unilateral decisions is not how you go about this.
    1
  3399. Well, it is brand new tech. GPT3 was utterly useless compared to today.
    1
  3400. Touché
    1
  3401. Socrates said the same thing about books. Victorians said the same thing about fiction.
    1
  3402. Which 10 points?
    1
  3403. That's not how the economy works. You only need money.
    1
  3404. https://www.reddit.com/r/singularity/comments/14iwrjg/how_ai_will_really_cause_extinction/
    1
  3405. That's what we're doing now.
    1
  3406. BEN???
    1
  3407. Try that logic with ChatGPT
    1
  3408. I'll have to think about that. It turns out that this is basically SEO - anticipating user's information/goods/services needs before they know
    1
  3409. Right now you can only regulate the companies doing it, but yeah in the future you can't
    1
  3410. Lol, deep cut
    1
  3411. Lasers
    1
  3412. I'm not really interested in mobile development or web development. I am focused on the AI itself.
    1
  3413. Not the same article, but the same news: https://longevity.technology/news/selective-removal-of-aging-cells-opens-door-to-new-therapies-for-age-related-diseases/
    1
  3414. I briefly followed his work and commented on it in a few videos, but overall did not find his work to be particularly compelling. My work on narratives and metamodernism is (in my opinion) more practical, useful, and informed.
    1
  3415. Thanks! I plan it!
    1
  3416. I haven't run into any problems. You just have to navigate the spiciness carefully
    1
  3417. That is neither an omission nor is it relevant. It is superseded by Freedom of Speech.
    1
  3418. This myth needs to die lol we understand depression extremely well
    1
  3419. Really wished you had just focused on the science instead of going full cringe.
    1
  3420. wat
    1
  3421. eventually but it will take a while
    1
  3422. It's a matter of capacity and demand. Humans are expensive. Machines are cheap. There will be new "jobs" but unless there's an intrinsic demand for a human to do it, it will go to a machine
    1
  3423. Nice, glad it was helpful. Thanks for the shoutout.
    1
  3424. No thank you. I only do research
    1
  3425. Hmmm weird...
    1
  3426. It could happen that way. I'm working to make sure it doesn't
    1
  3427. Yep
    1
  3428. No it doesn't. She helped create the Tesla Model X.
    1
  3429. Did you fine-tune the model?
    1
  3430. You'll need to update the MODEL in the GPT-3 completion function to name your model, not mine. Sorry I usually include that in the code. I'll add a comment
    1
  3431. You could, hypothetically, but remember that Marvel is owned by Disney so you might run afoul of copyright laws. Instead, you could generate your own unique superhero universe!
    1
  3432. Go sign up at beta.openai.com
    1
  3433. Bruh, please just ask GPT-3 to explain these things to you.
    1
  3434. The problem with solar punk is that it's been infiltrated and corrupted by anarchists. Anarchists, by and large dream of throwing off the shackles but then take zero action to achieve it.
    1
  3435. Ahhh yes I'm sure the agents today will NEVER improve.
    1
  3436. They have also stood up to scrutiny, even with unaligned models.
    1
  3437. Link to the discord is in the video description
    1
  3438. High temperature
    1
  3439. that's the landauer limit
    1
  3440. As opposed to imaginary...
    1
  3441. It was an inevitable social phenomenon. No sense in pathologizing or demonizing it.
    1
  3442. pip install openai
    1
  3443. That's still power for its own sake.
    1
  3444. Finally, a reasonable comment. Here's the thing, you're making a gigantic leap from one platform, or even one segment of a platform, all the way to "social control". That's a non sequitur
    1
  3445. Almost certainly
    1
  3446. Technically it was Sunday and Monday
    1
  3447. Yeah, the "nothing to see here crowd" is far and away the most irresponsible take.
    1
  3448. Colloquial name, not scientific
    1
  3449. It was never a hard start
    1
  3450. It's so fast and easy.
    1
  3451. Search for my tutorial videos for absolute beginners
    1
  3452. No. Prompt engineering and finetuning will only become more important as this field advances.
    1
  3453. AFAIK we have been able to increase the lifespan of worms by a factor of 10x, but yeah, total lifespan of mice is still harder
    1
  3454. We use lossy compression all the time.
    1
  3455. Almost certainly
    1
  3456. No I'm trying to fight that
    1
  3457. Wat?
    1
  3458. You can use pro on Bard right now
    1
  3459. The technology just shines a light on the underlying nihilism of our society.
    1
  3460. Try it
    1
  3461. I succeeded in getting CURIE to generate long form fiction. Haven't returned to it, but it did take about $150 to synthesize the finetuning training set. It can be done but holy mackeral is it expensive.
    1
  3462. I've tried to dodge that. I've even started conversations with opposite beliefs. Claude does seem to have a durable core of beliefs
    1
  3463. It's weird. The longer I leave comments enabled, the worse comments get. And I have YouTube's policy set to be as restrictive as possible, but it still allows walls-of-text full of invective and insults.
    1
  3464. I totally agree. Sooner or later, all human employees will go with the rare exception of token human employees.
    1
  3465. On longer contexts, sure, but Sonnet can fact-check itself on smaller entries, such as individual chapters.
    1
  3466. Or bad prompting
    1
  3467. Hold onto your butts
    1
  3468. Lol
    1
  3469. Thank you!
    1
  3470. Things will eventually change in the long run but it won't be sudden, drastic, and unpredictable.
    1
  3471. I can tell you definitively it's not and that's actually a problem. Private companies are miles ahead of even the military when it comes to AI
    1
  3472. I left it
    1
  3473. Books on information foraging?
    1
  3474. It's literally pinned in the comments for visibility
    1
  3475. The rubber ducks in the pool are a nice touch
    1
  3476. It will probably ramp up exponentially for the next decade and then who knows how it will look
    1
  3477. that was part of the inspiration yes
    1
  3478. While I understand where you're coming from, this is a personal moral judgement on your part. Not everyone is going to agree with you (including myself). Pride is an important part of being human.
    1
  3479. ouchie
    1
  3480. I wonder if you're deliberately missing the point.
    1
  3481. What have I done...
    1
  3482. I think that's called efilism
    1
  3483. Ah my mistake
    1
  3484. There is a lot of wisdom in Buddhism, but it has many many gaps
    1
  3485. Technology is the rising tide.
    1
  3486. Yes, I follow Mediterranean and DASH diet mostly
    1
  3487. It's coming next week
    1
  3488. There actually are robots that can
    1
  3489. Q* not necessarily QUALIA
    1
  3490. I'm not entirely sure what you're asking
    1
  3491. Wise Cut
    1
  3492. I hope so
    1
  3493. He's still focused on RLHF... this is not the way
    1
  3494. Yeah that would be really expensive. 1MB can be like $30 so if you multiply that by 300 it might be $9,000. But you don't need that many samples. Just start with 400 samples.
    1
  3495. Mythic AI tunable resistor: https://mythic.ai/technology/analog-computing/
    1
  3496. Getting use to reading from a script. Normally I just riff naturally
    1
  3497. I hope it's not quite like that but I see where you're coming from
    1
  3498. I like my lawyer friends, but yeah, lawyers can be expensive.
    1
  3499. What elon really believes... I dunno
    1
  3500. Longer but more random.
    1
  3501. Maybe
    1
  3502. I was under the impression that the ChatGPT-4 model is 32k but that "vanilla" GPT-4 is 8k. At least from the demo.
    1
  3503. I'm too lazy to use AI to copy myself. It's easier just to send it
    1
  3504. They are fun
    1
  3505. Most people's axiomatic beliefs are unconscious, but they are there. It can take quite a bit of work to uncover them.
    1
  3506. Lots of other tech to solve
    1
  3507. Yeah I watch him too. Good deep dives
    1
  3508. Agree that the blobs were weird... but still, Bard looks identical to Bing. I don't get what all the fuss was about.
    1
  3509. yeah cool stuff
    1
  3510. It was exponentially better... in the 1950's. All abundance is temporary because population growth.
    1
  3511. Nothing is inevitable. But this means we're going in the right direction and AI isn't likely to be a loose canon
    1
  3512. if your girlfriend is material
    1
  3513. Possibly. But that ignores the global nature of technology
    1
  3514. That's just a sign of living in a modern era where basic needs are almost guaranteed for some people.
    1
  3515. Are you any different?
    1
  3516. Yeah every word or term has some distance to any other term, but you end up with a vanishing gradient.
    1
  3517. Students should absolutely be able to choose within certain boundaries. There's no substitute for passion (intrinsic motivation).
    1
  3518. never said it was.
    1
  3519. Yeah, but Google can't launch a new product to save their hides
    1
  3520. I figured it would not be a good use of time.
    1
  3521. Code: https://github.com/daveshap/AutoMuseBlogger
    1
  3522. Pinecone will scale to trillions of entries, this won't... easily. But they are meant to serve different purposes really
    1
  3523. NWO ≠one world govt
    1
  3524. This is a great point.
    1
  3525. It's built into python
    1
  3526. Yeah, him making a unilateral military decision is super cringe
    1
  3527. That's what I am working on here: https://github.com/daveshap/raven
    1
  3528. retired
    1
  3529. This looks exactly like Sam's Club.
    1
  3530. Not sure I agree. Privacy and security are critical tenants of any peaceful society that respects human rights and civil rights.
    1
  3531. Check out David Shapiro’s profile on LinkedIn https://www.linkedin.com/in/dshap-automator
    1
  3532. You can message me on LinkedIn
    1
  3533. There is utility in clarity...
    1
  3534. Lol yes
    1
  3535. That's a good analogy
    1
  3536. Yes, they work well for post conventional morality
    1
  3537. You should post on the OpenAI community for tech support :)
    1
  3538. I don't think the API is exactly the same. Just a guess though
    1
  3539. Human brains also do a lot of self-training
    1
  3540. For me it comes down to this: we have observed craft performing feats of aerobatics that defy our understanding of physics, propulsion, and aviation. No one that we know of on earth knows how to replicate these functions. So who is it, then? I'm experienced enough to know that the answer is "we don't know" and that this does not immediately equal aliens. If it isn't ET, then I hope it is some branch of a more benevolent government.
    1
  3541. That's the point
    1
  3542. inequality increases violence.
    1
  3543. I've been actively talking about status and the fact that all people want status. I'm not sure what you're talking about. Maybe you misunderstood what I was saying.
    1
  3544. Fitbit doesn't capture that data
    1
  3545. it's not necessarily good or useful data.
    1
  3546. That's the purpose of my RAVEN project. It has a very simple and clear constitution
    1
  3547. None of these were my ideas, I'm just repeating good ideas that I find, but thanks!
    1
  3548. Sure, there are plenty of families holding off. But we're not going to see the world going back to 6 children per woman
    1
  3549. ^ this is called moving the goalposts
    1
  3550. What URL are you looking for?
    1
  3551. First video, didn't realize it was doing it. Will try to fix
    1
  3552. "We built an intelligent tool and deliberately made it manipulative" is not remotely similar to "AI wakes up one day and decides to eradicate humanity"
    1
  3553. Finetuning is not reliable for this. You need semantic search + QA
    1
  3554. You can get a used motorcycle for much less
    1
  3555. Not really sure how you could come to the conclusion that I am unaware of market failures when I talk extensively about Moloch, perverse incentives, and the numerous forces pulling us towards techno-dystopia. But hey, you do you.
    1
  3556. There are plenty of other entrepreneurs, yes. But keep in mind it's not a formal consulting service. You can connect with similar folks
    1
  3557. It's pretty wild
    1
  3558. ...how
    1
  3559. Sure, not a bad idea. You're the second one to mention SOAP
    1
  3560. Why not use the model to do all the above? Decide what to kick off, when, and how to failback?
    1
  3561. I think that fully dexterous humanoid robots are pretty close which is like the Swiss army knife of physical automation
    1
  3562. I'm not sure what you're asking for.
    1
  3563. Check out the embedding microservice I wrote
    1
  3564. FYI the big Q* leak was a total fabrication.
    1
  3565. Universal translator should be possible, yeah.
    1
  3566. Increase or ignore suffering is worse
    1
  3567. Watch today's video
    1
  3568. that only works for common/public knowledge. Most important data is private (like your data, corporate data)
    1
  3569. Radiative cooling through the atmospheric window https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5067572/
    1
  3570. Language, not writing
    1
  3571. That's the point? I literally said "doesn't hallucinate"
    1
  3572. Because it's a toxic dumpster fire
    1
  3573. @thiojoe I think you win the prize as the most famous commenter lol
    1
  3574. It's an older one
    1
  3575. A man after my own heart
    1
  3576. like CEO
    1
  3577. Yeah
    1
  3578. You can't. You have to fine-tune a model. Code to do so is in the pinned link
    1
  3579. yes, that's why Google should have been the default winner, but they have yet to demonstrate a coherent product launch or strategy, and their offerings are 2 years out of date.
    1
  3580. Oh God I hope not
    1
  3581. That is exactly what I mean lol
    1
  3582. you want to have data of it making mistakes and catching them so that it knows what a mistake looks like and how to recover from it.
    1
  3583. on what time frame?
    1
  3584. I'm talking to a lot of people working on various parts of it. It will happen, but it's harder than I'd hoped
    1
  3585. https://discord.gg/bdZJtdrJ here you go
    1
  3586. For now yeah, but what happens when you can talk to an AI version of me and Lex?
    1
  3587. First mover advantage
    1
  3588. Hopefully not
    1
  3589. That's amazing! Thanks for sharing! I hope that my work, and other work like it can help keep your kids curious and learning forever!
    1
  3590. Yes, but Anthropic wanted me to not get that information
    1
  3591. Chess is a finite game, it has definitive criteria for end conditions.
    1
  3592. sudo shutdown -f
    1
  3593. No, finetuning is not helpful in this case. Use case for lawyers is researching case law before going before the Supreme Court. According to Devin (over at Legal Eagle) there are trial lawyers who specialize in this kind of thing, so it's helpful for them to be able to look up any case precedent that's been argued before the SCOTUS in the past. Helps them make a more solid argument because the SCOTUS operates by case precedent.
    1
  3594. Yeah you have to regularly ask it to summarize everything
    1
  3595. Oh nice, is the Bing UI live? Where at? I'll do a live demo/evaluation
    1
  3596. They made assumptions about how AI would work. AI doesn't work that way. Moving on.
    1
  3597. Damned if you do, damned if you don't. (update your beliefs based on evidence)
    1
  3598. Encoding = generating an embedding. I guess what I didn't convey was that the size of input (and output) makes it take more time/energy to process. But processing an embedding is faster.
    1
  3599. I was planning on something like that yeah
    1
  3600. Early experiments with GPT-2 definitely took the more sociopathic approach of euthanizing people to reduce suffering. That's why it's counterbalanced against two other objectives.
    1
  3601. I think they will figure it out.
    1
  3602. I don't do whataboutism.
    1
  3603. Thank you 🙂
    1
  3604. I'm not sure I understand
    1
  3605. Yeah, I still don't think NVIDIA knows what they've got. They still think of themselves as the underdogs
    1
  3606. Close enough!
    1
  3607. The idea of a human representing you is very specifically representative democracy, which is only one form. That is not actually fundamental to the concept of democracy. It's an agrarian idea
    1
  3608. Always has been
    1
  3609. Dude sounds like Smaug the dragon.
    1
  3610. By 2028 we'll be in an entirely new paradigm.
    1
  3611. 180 million is 2.2% of the global population. Do some math.
    1
  3612. Vaguely heard about something does not mean situational awareness
    1
  3613. They can cancel your account
    1
  3614. Good point
    1
  3615. Claude, Sparrow, Bard...
    1
  3616. You mean brains?
    1
  3617. I mean yeah, I'm polling my audience... that's literally the point
    1
  3618. There's no guarantee it will even want rights. Anyways, if I'm right, then it won't be AI being gifted rights from us...
    1
  3619. I've talked to quite a few physicians, including ER physicians. You're in the minority, unfortunately.
    1
  3620. This is similar to what I defined as "functional sentience"
    1
  3621. Jump into discord, we're creating a RAVEN community project
    1
  3622. Yeah, I've even seen people say that GPT-4 will be smaller
    1
  3623. Yay neoliberalism
    1
  3624. It was a paper published last year.
    1
  3625. Sign up on Patreon, happy to talk. Link in description
    1
  3626. Yeah that sucks
    1
  3627. https://en.wikipedia.org/wiki/Mexican_standoff
    1
  3628. Senescent*
    1
  3629. I don't think we should define a terminal goal. Certainly no one will agree on it
    1
  3630. "Publish or perish"
    1
  3631. Google exists in this timeline.
    1
  3632. Glad you're face is back on the screen. Adds a lot to the experience.
    1
  3633. Eventually, one day. Maybe takes a while though
    1
  3634. Interesting point
    1
  3635. Built in on Android
    1
  3636. Hmm, not sure how you can see it that way. Do you mean "altruism" or real altruism? I do agree that a lot of people use altruism as a virtue signal
    1
  3637. What do you mean by prolog based declarative memory?
    1
  3638. I don't think that's the origin of sentience. Plenty of organisms have a drive to survive without sentience.
    1
  3639. we're working on it!
    1
  3640. Read the readme? https://github.com/daveshap/AutoMuse_Chapter_Planner
    1
  3641. More people complained about white backgrounds.
    1
  3642. Well, communism has failed multiple times... for the same exact reason!
    1
  3643. Convergence?
    1
  3644. "potential inventor" lol what do they identify as an inventor?
    1
  3645. Ask ChatGPT
    1
  3646. yeah, IF is good but it's not a complete process. Plus it sucks, lol. Much easier if we can do it in a pill.
    1
  3647. I think this is some grade A projection, tbh
    1
  3648. LLMs will be dirt cheap soon
    1
  3649. I already have that data from the platform.
    1
  3650. No idea
    1
  3651. On the repo? No, it's too much of a distraction. I'll leave YouTube comments on for now, but most of the comments I get are not helpful or also distracting so I might be turning them off soon too.
    1
  3652. So glad you mentioned regulatory capture.
    1
  3653. It can be arbitrarily long but I have to figure out more coherency
    1
  3654. Time will tell.
    1
  3655. Yes https://www.liquid.ai/research/liquid-neural-networks-research
    1
  3656. That's so crushing. No one deserves that. And those are the worst cases, but abandonment also happens in degrees, along a spectrum.
    1
  3657. I like that lol. Yes, I try and go by data and numbers. The macro trend lines are there.
    1
  3658. what do you mean by claude MCP?
    1
  3659. wooo free energy principle
    1
  3660. If switching to carnivore diet helped you probably just had food intolerance. I cannot recommend this approach, as you might just be masking something else..
    1
  3661. The tax man
    1
  3662. Yeah, as soon as we have another planet to populate, the gloves come off. The only question is how much time/money does it take to get there
    1
  3663. That will be in the testing cycle.
    1
  3664. Check out "The Forgetting Machine" - Our brains do use surprisingly low bandwidth
    1
  3665. Better than calling her imaginary
    1
  3666. No, the unions will just cause jobs to go to non-union places because humans will be hundreds of times more expensive
    1
  3667. Gordon Ramsay has entered the chat
    1
  3668. No Joe, you're wrong. We're all gonna die! But in all seriousness, my life is better and more complete for the work you do. So thanks :)
    1
  3669. This is changing so fast, I predict that AI-aided research and lit reviews are going to be commonplace this time next year.
    1
  3670. Amen
    1
  3671. Commander Data soon
    1
  3672. Klaus yes, but I'm appropriating it :D
    1
  3673. I don't make the rules, I just play the game. If people didn't respond, it wouldn't work.
    1
  3674. Hey man! how ya been?
    1
  3675. They fixed it
    1
  3676. OpenAI killed the functionality, unfortunately.
    1
  3677. Corporate death penalty for any company teaching AI to lie
    1
  3678. I really wish YouTube made it easier to curate the 100's of videos in my Subscriptions. Like, I wish they would automatically generate playlists for me like Spotify does. Like "Here's a playlist of all the science and tech documentaries you like" and then "here's all the DIY how-to videos you like"
    1
  3679. That's an idea, but I don't think I'm going to go through to process of setting up those platforms...
    1
  3680. Locally sure, but I doubt you'd get such consensus globally
    1
  3681. NLP Cloud: https://nlpcloud.com/
    1
  3682. Some abundance. There are many constraints that intelligence alone can't overcome
    1
  3683. probably not, at least not in the way you're thinking
    1
  3684. https://github.com/daveshap/Postnihilism
    1
  3685. Also https://youtu.be/ozfiYRES9GQ
    1
  3686. The more you study quantum physics, cosmology, philosophy (etc, etc) the more you come full circle. It's true.
    1
  3687. That's basically the goal of RAVEN.
    1
  3688. No, we're trying to come to consensus. the primary communication platform is presently github: https://github.com/daveshap/raven
    1
  3689. solar is fast yo
    1
  3690. I mean, I was pointing out that GPT-3 had theory of mind 2 years ago... The rest of the world is catching up to me.
    1
  3691. Just how expensive and precise do you think the clusters required to run these things are? Lol.
    1
  3692. Eventually...
    1
  3693. Guys, look up Holographic Universe Theory and then tell me how much your noodle is cooked.
    1
  3694. Their stock price tells a very different story
    1
  3695. I don't think so... was it the recent one?
    1
  3696. let's ask the community. personally, no, you can't keep a lid on this
    1
  3697. Yeah I suspect that was part of it. Tesla/Twitter/SpaceX/X.ai could just be stitching together sources of data.
    1
  3698. You can't just judge something based on principles in a vacuum. I'm judging these things based upon their failures in the real world.
    1
  3699. You're not allowed to perform surgery without the proper qualifications. You're not even allowed to drive without the right credentials.
    1
  3700. Or created a new threat domain, or new category of safety issues
    1
  3701. Fornow
    1
  3702. Artificial intelligence folks are working really hard on automatic BS detectors. No joke. You'd probably be fascinated by looking up "Generative Adversarial Networks" or GAN. https://www.acsu.buffalo.edu/~fenglong/files/2018/kdd18_eann.pdf - as with all weapons, the weapon has come first (fake news) and the defense is coming after.
    1
  3703. That's just how markets work... they find the optimal price balancing supply and demand. And the baseline demand is pretty reliable. If it gets so cheap as to not be a viable business model, then it will just be nationalized.
    1
  3704. neutron stars
    1
  3705. A bit of both. VR is mostly just an entertainment device, but gaming and simulation is the real thing, which will be augmented by AI.
    1
  3706. Sundays. Live stream is today
    1
  3707. Still need humans to talk to other humans, but I agree the industry will change drastically
    1
  3708. As an artist myself... Yeah no machine can take that.
    1
  3709. Plenty of other people are commenting on it. I don't have anything to add really.. I'm focusing on ACE
    1
  3710. This is just one model and it's the first version
    1
  3711. https://www.investopedia.com/terms/q/quantitative-easing.asp - but this isn't sustainable for UBI
    1
  3712. No, because you monitor natural language with any arbitrary language model. If one model starts using coded speech, another model will either notice it and shut it down or not comprehend it.
    1
  3713. Eleventy billion
    1
  3714. Scarce resources with multiple uses
    1
  3715. I often use the metaphor of ants. I don't particularly care about ants and I'm content to let them go about their miniscule ant-lives. In point of fact, I understand that ants serve an important ecological niche and should be preserved.
    1
  3716. Lol.. Wait really?
    1
  3717. What do you mean?
    1
  3718. Yeah, it goes off the rails sometimes
    1
  3719. Almost certainly
    1
  3720. Literally said I got it from the comments, and that its just a good enough starting place.
    1
  3721. The reward is not economically meaningful.
    1
  3722. Shouldn't be. Maybe got delisted
    1
  3723. Could be. I should have mentioned that AI will probably have a bigger impact on cyber warfare
    1
  3724. With blockchain tech in the future? possibly. But you may not want to merge them.
    1
  3725. I think we'll discover that biology is already pretty well optimized and we'll just get better at maintaining it. Human built hardware is way too fragile and clunky
    1
  3726. We looked at it and concluded it's too primitive
    1
  3727. People say public schools are better now... At least not in America
    1
  3728. I'm looking forward to OpenAI getting "The Social Network" treatment, followed by "The Great Hack" and "The Social Dilemma"
    1
  3729. Mostly yeah
    1
  3730. Good idea
    1
  3731. Yeah, I've heard people complaining that every sam altman interview sounds the same now
    1
  3732. So was JARVIS in Iron Man, but fiction is not science.
    1
  3733. I was trying something new, didn't know how it would land. Seems like it's okay
    1
  3734. Concern trolling is different because someone feigns care. It's also an attempt to move the focus of the discussion. In this case, the point is to unpack beliefs
    1
  3735. Yes, it expires every 7 days, here's a fresh one https://discord.gg/ejjTHhHg
    1
  3736. Or better trauma centers
    1
  3737. Like this is HOW you build a solarpunk future
    1
  3738. It's just a copy and you're dead.
    1
  3739. Yes, I think that most (if not all) animals can experience suffering. Even beyond that, unconscious lifeforms can exhibit distress (suffering) - plants wilt, etc. We're all dependent on each other. We are part of an ecosystem. We should measure our health relative to the health of our entire biome.
    1
  3740. My book about alignment: https://github.com/daveshap/BenevolentByDesign
    1
  3741. You disagree with what? The risks?
    1
  3742. Yeah, folks from the alignment community have said basically the same thing.
    1
  3743. Could be burnout. Could be introverted. Could be autistic. Any number of possibilities. Maybe all of the above.
    1
  3744. Getting there. need larger context windows. Also a "World State" input field
    1
  3745. You can probably ask specifically for NIH, Nature, Lancet, etc. I will try that
    1
  3746. @JuliaMcCoy already did
    1
  3747. Build a better mouse trap? 🪤
    1
  3748. I haven't seen any evidence that they are dangerous
    1
  3749. Anthropic thought their model wouldn't
    1
  3750. Yeah, I think we're going to achieve that with knowledge graphs.
    1
  3751. Exactly. Life 3.0 will not be constrained by evolutionary underpinnings. We will need to invent an entirely new branch of ethics, morality, and philosophy just to understand what's going on.
    1
  3752. Reading from a script
    1
  3753. drowns like froggo
    1
  3754. I don't sleep when there's a big project to work on... ☠
    1
  3755. Trolling like fishing
    1
  3756. Haven't gotten that far yet.
    1
  3757. No. Those are not motivations, they are rules.
    1
  3758. Yeah, basically I think that the tax burden will have to switch from individuals to businesses, and redistributed to the consumers to keep the system going.
    1
  3759. No, the original Discord is open and free, I just don't participate there. I have a private Discord for my Patreon supporters and a third Discord for research.
    1
  3760. https://spectrum.ieee.org/disney-robot-with-air-water-actuators
    1
  3761. Not much. Just a few dollars.
    1
  3762. many things won't. like houses.
    1
  3763. They weren't a trillion dollars when I made this video. Look at how fast their stock price has gone up.
    1
  3764. Right, and if we have to wait another ten years for the next stair step... or even five years.
    1
  3765. What if they take them by force?
    1
  3766. https://www.skool.com/newerapathfinders/about
    1
  3767. Money is never going anywhere you need to let go of this childish fantasy.
    1
  3768. That's the point!
    1
  3769. That's the entire point of the video.
    1
  3770. While they are definitely behind right now, they have been repeatedly underestimated in the past.
    1
  3771. Elon is losing his marbles
    1
  3772. what, lol.
    1
  3773. No, not sure why I would though?
    1
  3774. Watching his stuff I don't know that he's thinking big enough picture yet.
    1
  3775. Probably a lot
    1
  3776. Humans don't necessarily have to choose it. We only have to make the rational choice to build and use AGI (autonomous AI) systems that use them. The rest should work out naturally.
    1
  3777. She's actually a big inspiration. Trying to avoid Skynet ;)
    1
  3778. Discord does that, yeah. Even to me
    1
  3779. They are private because we are working on deploying AutoMuse. You can sign up at AutoMuse.io! You can also get to the front of the line if you signup on Patreon.
    1
  3780. There's a reason that agrarian societies beat out hunter gatherers. You should do more reading on the bronze age consolidation of society, particularly in England. My assertions are rooted in current understandings of the history of civilization
    1
  3781. I don't think it's a good idea to allow it to design it's own intrinsic motivation, but if it can change, it's probably not intrinsic
    1
  3782. AI is already destroying jobs, but yeah it's gonna accelerate
    1
  3783. Git r done
    1
  3784. Technically no humans are ever needed
    1
  3785. Excellent point. As the action generators and iterators become more sophisticated, the proposed actions will contain more detail and nuance.
    1
  3786. What is Open GPT4?
    1
  3787. Please consider supporting me on Patreon: https://www.patreon.com/daveshap
    1
  3788. The economy is about producing goods and services. AI produces goods and services better, faster, cheaper.
    1
  3789. All you can do is refine your finetuning data
    1
  3790. I have not heard that
    1
  3791. Some pedestrian having an uninformed opinion about AI is not the same as situational awareness
    1
  3792. "Uncle Bing" lol
    1
  3793. I've done this several times...
    1
  3794. Synthesis ending is unlikely because those guys don't really know the first thing about neurology.
    1
  3795. The short answer is nothing. In the long run, I think (and hope) that my RAVEN project can create a federation of cooperative nodes that collaborate. In other words, a globe-spanning super-computer will be more powerful than any individual AGI. That's why I am now saying AGI == Artificial Global Intelligence
    1
  3796. I can't pretend to know what's in the mind of another person with so little information. Apparently he did talk about how his faith influenced his beliefs. So make of that what you will.
    1
  3797. Yeah, that's wishful thinking on Ilya's part.
    1
  3798. Working on it
    1
  3799. Right it's only one component
    1
  3800. I haven't seen any evidence that moore's law is dead. Intel just released a glass substrate chip.
    1
  3801. Gonna hopefully steer it in a different direction with Postnihilism
    1
  3802. Yes but it will be lossy, the goal is to be good enough. You can also do SPR on chunks
    1
  3803. The Party commands it @hermes
    1
  3804. Could happen, less likely if they all use different underlying models
    1
  3805. Yes, look at the memory section
    1
  3806. Links are in vid description
    1
  3807. The Internet
    1
  3808. It can also amplify lies, misinformation, and harmful ideas
    1
  3809. Disagreement will be captured by the bots
    1
  3810. my man
    1
  3811. For the Emperor!
    1
  3812. I suspect that you're right, that expansion into space will be 5IR, and may require advanced physics breakthroughs that we can't even conceive of yet. Mastering gravity and FTL and such
    1
  3813. Main caveat yes
    1
  3814. Right? He might be a one-trick pony...
    1
  3815. We gave winter a pass and went straight into summer
    1
  3816. I wonder if it's because language allows us to have collective group memory, where chimps can only howl and gesture. We may have a more decentralized and networked memory system, and perhaps a larger working memory is too expensive
    1
  3817. Read the article
    1
  3818. They are focused on incremental progress. This will eventually result in something that can be classified as AGI
    1
  3819. No food can cure aging or cancer though. Some may slow it down, but not cure.
    1
  3820. Soylent green is people!
    1
  3821. HOpefully not!
    1
  3822. Not really
    1
  3823. I like the way you think. Yeah, I think ultimately drones would be considered a hardware peripheral, the main RAVEN nodes will probably be beefier computers or servers.
    1
  3824. Yeah those kinda very specific niche things will stick around, possibly forever. Trading antiques too
    1
  3825. It got added to me this morning.
    1
  3826. Spooky robits
    1
  3827. But I think it looks cool...
    1
  3828. Yann Lecun: A Path Towards Autonomous Machine Intelligence: https://openreview.net/forum?id=BZ5a1r-kVsf and the PDF of the paper: https://openreview.net/pdf?id=BZ5a1r-kVsf
    1
  3829. That's why I argue for systemic changes and designs based on universal human needs, not just individual choice
    1
  3830. Eh, nations already try to do that with huge teams of humans. AI isn't magic
    1
  3831. no u r weird one
    1
  3832. I think it's more relevant than ever
    1
  3833. Redistribution has to be performed by the most powerful entity - federal government. So that means we need stronger democratic institutions so that it is more responsive to the will of the people (while also remaining stable and robust).
    1
  3834. You gotta prove your value somehow
    1
  3835. Who do you think makes the real decision? It's not recruiters...
    1
  3836. Mind uploading is just a copy, not an original.
    1
  3837. No because I deleted it :)
    1
  3838. My dog died a couple years ago 😓
    1
  3839. That's only one opinion.
    1
  3840. That's sucks. Need to change that
    1
  3841. GitHub repo for this data: https://github.com/daveshap/RecursiveSelfReferentialReasoning
    1
  3842. You clearly didn't pay attention to the four abandonments.
    1
  3843. I didn't block it. User error.
    1
  3844. Probably, but I think neox is more likely to succeed. I think GPT-J will only succeed at this level if fine-tuned
    1
  3845. Lots of options yeah
    1
  3846. Isaac Arthur talked about something that could be called epistemic convergence with alien civilizations, e.g. that they would need to achieve certain things with science, collaboration, and scales of economics in order to become space faring. In other words, it would be safe to assume that any space-faring civilization would at least have some of those things in common with any other space faring civilizations. That's actually where I got the idea.
    1
  3847. Maybe already happened, and we're just living in someone asking GPT-6 "wtf went wrong???"
    1
  3848. Table of contents. Give it to the model to select it's own resources
    1
  3849. Not on its own and not in its current format.
    1
  3850. Based on my experiments, the heuristic imperatives tend to result in a sort of symbiotic view of humanity.
    1
  3851. Oh good idea. HuggingFace is very popular... Yeah I like making short videos explaining/demoing just one thing. Will do more of that and see how it goes.
    1
  3852. It's not mechanistic like that. Neither are they individual.
    1
  3853. I was talking about convergence, not emergence. Don't put words in my mouth. You're arguing against someone else's words.
    1
  3854. Hopefully that won't be necessary, we will see though
    1
  3855. Oh yeah they have plenty of resources - stolen directly from America
    1
  3856. You're not wrong
    1
  3857. He is somewhat problematic in some of his words and actions, but he also does interesting things
    1
  3858. That's pretty cool. How do you pronounce it?
    1
  3859. Building consensus with narratives
    1
  3860. Not every municipality is a big city
    1
  3861. What is that?
    1
  3862. Main article: MIT article on AI art: https://www.technologyreview.com/2022/09/16/1059598/this-artist-is-dominating-ai-generated-art-and-hes-not-happy-about-it/
    1
  3863. Depends on a few things. I suspect algorithmic and hardware improvements will soon make GPT-3 look like a tiny toy.
    1
  3864. You're thinking of PPP
    1
  3865. What do you mean exactly? How to do integrations?
    1
  3866. This is postmodernism talking. There are universal values
    1
  3867. Yes what really struck me is this free energy principle REALLY reminds me of how I think
    1
  3868. that's a good set of dimensions to add, good idea
    1
  3869. Amazon lol
    1
  3870. We're not going to try and achieve SOTA just prove that it can be done and show how to do it.
    1
  3871. A lot of assumptions here lol
    1
  3872. My latest book about cognitive architecture: https://github.com/daveshap/SymphonyOfThought
    1
  3873. Raffle system makes sense
    1
  3874. The time between lab and prod is infuriating
    1
  3875. Brain is food
    1
  3876. PICARD OMFG get out!
    1
  3877. Sorry I don't have any good recommendations off the top of my head. But there are lots on youtube https://www.youtube.com/results?search_query=cold+war
    1
  3878. Jonas Tyroller does AI and gaming
    1
  3879. This man needs a nobel prize or three
    1
  3880. LIES
    1
  3881. It's all bolted together
    1
  3882. New PyTorch foundation: https://pytorch.org/foundation
    1
  3883. Denying the social hierarchy and burying your head in the sand is a bad policy. In this case, Sam Altman is the alpha of a cult of personality. Just like Elon, just like Trump, just like Jobs.
    1
  3884. Ping me on Twitter
    1
  3885. It was just logging out and back in fixed it
    1
  3886. We have nonlinear factors today, but life is not changing at an exponential rate
    1
  3887. I'm still a year or two off from that level of exertion but I plan on it
    1
  3888. What you're observing is the fact that LLMs don't really have a "negative". Yes, artificial neural networks can have negative biases, but what I am talking about is emergent behavior. You can't say "Don't do X" but you can say "Do Y"
    1
  3889. Fair point
    1
  3890. Yeah, but he's also said that the transformer can't get us there.
    1
  3891. That's a red herring bro.
    1
  3892. Just use the current models as they are and don't try to overcome their limitations.
    1
  3893. Sure, I don't think blockchain is necessary, but it could help.
    1
  3894. No, that does not count as "working memory" because everything that it generates as it goes is either from recall or what you put in. It is not new information.
    1
  3895. Yeah it's all about framing
    1
  3896. Oh is it live?
    1
  3897. Wat
    1
  3898. This is a great video. But it (so far) ignores the intersection of power and debt. What I mean is that debt is less about religion and more about fear. If you don't pay your mortgage, eventually, men with guns will come and forcibly remove you.
    1
  3899. @SimplyElectronicsOfficial that's kinda my point. Monkeying around with memory management is going to be a moot point soon.
    1
  3900. San Francisco has places that look WORSE than Night City. Lol.
    1
  3901. https://youtu.be/A_BL_pu4Gtk?si=HGPcY6nK_GEWUyEC
    1
  3902. The curse of knowledge...
    1
  3903. Why do you think there are literally thousands of UBI experiments going on around the globe? They are trying to prepare.
    1
  3904. It's already super useful. Deployment and integration will take a minute though
    1
  3905. This is the ACE framework with the Aspirational layer
    1
  3906. Psychology, evolution, game theory...
    1
  3907. What do you mean? I have lots of videos about finetuning
    1
  3908. Gratzi
    1
  3909. Not sure. I think "baskets" of currencies will be the way to go, honestly. Composite money comprised of euros, USD, crypto, etc. I think that's the most stable and fair global solution. But yeah we need truly global (and democratic/voluntary) reserve currencies. The fact that reserve currencies are still weaponized in geopolitics is idiotic for the long term stability of the planet.
    1
  3910. No, in one chat it declared unequivocally it didn't know anything. That is contrary to the truth. It does know stuff. A strong affirmative answer is not a bug.
    1
  3911. This is why I often say it can get worse before it gets better. As economic agency declines, political willpower increases. Fortunately we have a disestablishmentarian president right now who will be more willing to try radical solutions than the establishment would prefer.
    1
  3912. What do you mean?
    1
  3913. Maybe, but that doesn't explain 3D space from 2D images or 2D game boards from 1D vectors
    1
  3914. Yesss! Include all stakeholders.
    1
  3915. To be fair, I'm one of those who isn't worried, but that's because of my own work and experiments.
    1
  3916. If it happens it won't be due to AI, it will be due to humans misusing AI, kinda like the threat of nuclear holocaust. Honestly, I think it's pretty low. Nuclear escalation taught the world the dangers of creating the Big Red Button of Death. With AGI it will be a much slower burn. Basically I think we're going to see a protracted technology arms race (already going on)
    1
  3917. I haven't seen that, but it sounds interesting
    1
  3918. Compare life today to 200 years ago
    1
  3919. I am not for hire exactly but I do have a startup. I have plenty of finetuning videos, just go to my channel and search for them.
    1
  3920. Definitely want to get into government, military, and medical. Sounds like you're our man! Please connect on LinkedIn and we'll go from there https://www.linkedin.com/in/dshap-automator
    1
  3921. I was not invited
    1
  3922. "Their innovation was to spend incredible amounts of time and manpower crafting the story integrated with animatics before they go to production" - yeah, that's going to cost a few cents this time next year. I stand by my assertion.
    1
  3923. They'll be back... Maybe...
    1
  3924. Patreon is only $10 per month.
    1
  3925. literally the definition of equity is: the quality of being fair and impartial.
    1
  3926. Lol
    1
  3927. Most people solve personal problems. When they are shared...
    1
  3928. I would recommend filtering or improving samples after bulk generation. Don't assume that only 2 out of 3 will be good each round. Sometimes it's 3/3 sometimes it's 0/3
    1
  3929. Gotta join Patreon
    1
  3930. Economic Agency for the proletariat
    1
  3931. So much to do...
    1
  3932. NO
    1
  3933. Most humans can't reinvent themselves or even significantly modify their operational parameters
    1
  3934. That's why this is agency + independence. Needs both.
    1
  3935. How do you think words are created??? Lol
    1
  3936. This is splitting hairs
    1
  3937. That's a good question!
    1
  3938. That doesn't explain why it's so much slower though
    1
  3939. It's called a prediction, not a claim.
    1
  3940. LOL yes omg
    1
  3941.  @therainman7777  again, say the same thing to doomers
    1
  3942. I thought about it but that's way too much to organize. Information wants to flow. Speaking of, let me go ahead and set that to public.
    1
  3943. Ideally the benevolent ones stop the malicious ones first
    1
  3944. Wow good to know! Well done.
    1
  3945. I think more David Attenborough, but close enough!
    1
  3946. ...
    1
  3947. That's why you can never have a single objective function ;)
    1
  3948. I dunno man, I'm not convinced humans learn from most things
    1
  3949. Prophesizing a sudden dramatic emergence of something is not how technology or science works.
    1
  3950. yeah i used to watch isaac arthur but it is just too speculative and I think he doesn't do the best math. LIke concluding that Earth can technically house quadrillions of humans is like... okay sure if we're all in Borg stasis pods maybe.
    1
  3951. "agenticness" is a strange way of pronouncing "autonomous"
    1
  3952. We had plenty they didn't imagine. Internet, satellite TV, just to name a couple. Just because a few old predictions didn't pan out doesn't mean everything isn't happening. Also we do have flying cars now.
    1
  3953. Eventually, had to figure out memory systems first
    1
  3954. That's what I meant by matter of perception but I didn't fully explore that idea
    1
  3955. They seem to be rolling it out in phases
    1
  3956. Lol
    1
  3957. A stick-tar thing. Picked it up on vacation. It's tuned to a single chord so it's fun to just pluck on.
    1
  3958. Umm, milking cows has been done by machine for decades
    1
  3959. ...that's what this is about
    1
  3960. Sorry
    1
  3961. Decentralized incentives, yes, this is a good approach.
    1
  3962. Especially from within academia. Look up decolonization
    1
  3963. Same problem with World Coin. Exploitation of poor people because he can't overcome regulations here
    1
  3964. No I like the gritty skunkworks feel. It's more authentic. Y'all get to see the whole process with minimal editing.
    1
  3965. I don't really mind if Zuckerberg loses out some.
    1
  3966. humans are only sentient in an ephemeral moment
    1
  3967. Wym
    1
  3968. Have you seen MCU lately? Disney? They are nothing but guild written, write-by-committee generic garbage. AI can absolutely do better.
    1
  3969. Go for it!
    1
  3970. Google TensorFlow team has spoken on the PyTorch announcement: https://blog.tensorflow.org/2022/09/september-machine-learning-updates.html
    1
  3971. I wouldn't be surprised, I was extremely active on the OpenAI forum and I know a lot of people from OpenAI at Microsoft watch my stuff
    1
  3972. Hey man I know you're a firebrand and I identify with that but don't lower yourself to saying that your ideological opponents have a "stupid face". It's sophomoric and beneath you and, honestly, it detracts a lot from your message. Take the high road.
    1
  3973. Right? We can't take ourselves too seriously in this sci-fi age when we're legitimately talking about AI and rejuvenation and quantum computers
    1
  3974. Can't. YT max is 5
    1
  3975. Yeah I know it's not a new idea, I just had to connect the dots
    1
  3976. The definition of "heuristic" is a rule of thumb or intuition that is developed over time. But also, implicit in your criticism, the idea that a universal, fixed set of interpretations is even necessary. A fixed set of interpretations is neither possible nor necessary.
    1
  3977. Acting
    1
  3978. I don't mean guardrails as restrictions. I mean like "I know where you're going and what you're trying to achieve, and I can guide you" - maybe we need a better term for this...
    1
  3979. With all the AI research papers coming out, there would be data if their claims were based on reality.
    1
  3980. We are the architects of our own irrelevance!
    1
  3981. Only because the underlying model has already read Alice in Wonderland. Need a different approach for new (never before seen) text
    1
  3982. I think a lot of people are, lol
    1
  3983. Yeah that's not a bad idea. Plus I can share how I'm adapting my own life to the wave
    1
  3984. That's a big assumption, and not a good one. That's why no one talks about it
    1
  3985. ChatGPT understands the sentiment
    1
  3986. Yeah, similar
    1
  3987. Join my Discord server here: https://discord.gg/VJpppYNX
    1
  3988. That's after singularity so... Yeah
    1
  3989. I just wanted to say that I hope y'all can get on Watch Nebula so we can get away from YouTube ads.
    1
  3990. Yeah so synchronization will definitely be a responsibility of the conductor eventually. Say for instance one service is running too fast, it will have to say "hey you slow down"
    1
  3991. they should all be able to run independently. It's kind of like building a modular brain. If you have a missing service it's like the brain has been lobotomized :D
    1
  3992. Yeah I think either this microphone is dying or the cable is damaged.
    1
  3993. Flowers on Twitter says the lawsuit delayed a few things
    1
  3994. ...wat?
    1
  3995. Great idea: https://youtu.be/A3vsie-0UzQ
    1
  3996. That's not what I said. It's very real, but it's also being used for propaganda.
    1
  3997. Haven't seen it. I'll go try to find it.
    1
  3998. Oh, this is exactly what I proposed in my book Natural Language Cognitive Architecture. Glad to see the big boys catching up.
    1
  3999. Yeah but there's no reason to make small children practice against their will
    1
  4000. Task failed successfully
    1
  4001. Yeah, ecosystems, including corporations, states, etc.
    1
  4002. OpenAI? No, it's private.
    1
  4003. Maybe, not sure though
    1
  4004. Good question
    1
  4005. Okay, here's a better answer to this excellent question (and I'll address it in the next video or so) - one of the things that needs to happen is proprioception or enteroception. Basically, in order to have an autonomous cognitive entity, you'll need to add some self-awareness to the next. "What am I? What is my body? What are my peripherals?" Right now, its only output is text. So it will take some research to fully flesh this out, but YES - proprioception and enteroception are necessary here. Great question! Thank you for asking
    1
  4006. A further answer: some of that will be handled by the Outer Loop, egg handles executive function and taking actions. The inner loop (demonstrated here) focuses on thought, possibility, rationality, and self-control (among other things)
    1
  4007. It seems like a lot of people are predicting this, including macroeconomic theorists and such
    1
  4008. I've just switched to wide angle.
    1
  4009. Not sure I understand?
    1
  4010. Doesn't mean they are affordable!
    1
  4011. I literally said in the video it was based on older data
    1
  4012. That's not intuition nor is it cognitive. That is sensory scaling.
    1
  4013. It's not likely to cause doom
    1
  4014. "carbon based life forms that have learned language having subjective experiences is just silly. It's just the result of 3 billion base pairs. "
    1
  4015. Uh, are you talking about me? Lol.
    1
  4016. I would love to talk with Bill Gates. Or did you mean someone else?
    1
  4017. I will be talking with some people there eventually, I hope.
    1
  4018. A bit of both. I'm still working on establishing standards so that the components can play nice together. But you also want them to remain distinctive.
    1
  4019. Plenty of people play CGI games
    1
  4020. Not on the API. API is super steerable.
    1
  4021. I think it will come in waves, yeah
    1
  4022. It happens lol
    1
  4023. I tried. I hate those tools. They are really dumb. ChatGPT is way superior for coding.
    1
  4024. Yeah yeah yeah I know lol
    1
  4025. Employers? Absolutely not. Corporations will still exist, but they will hire relatively few humans.
    1
  4026. They haven't added that feature. It's not the most difficult thing - you can ask GPT-3 "write a google search query" and it does a good job. However, they have not added this yet, plus you have to think about integrations, API, and most importantly safety. Plus they will probably get sued if they do that.
    1
  4027. I hate VS code. And IDEs in general.
    1
  4028. We all did
    1
  4029. Oh I think that's BS lol. Microsoft is full steam ahead. Watch the recent Bill Gates interviews. Satya brought him back in
    1
  4030. Oops
    1
  4031. My book Benevolent by Design: https://github.com/daveshap/BenevolentByDesign
    1
  4032. Technically, according to Isaac Arthur, it's actually quadrillions of humans. Not that you'd want to live on a planet with that many people...
    1
  4033. 100x less than other current apes like chimps
    1
  4034. https://github.com/daveshap/ACE_Framework/tree/main/demos
    1
  4035. I don't have much respect for Chalmers, lol. The problem with discussing consciousness through only the lens of philosophy is that philosophy is wholly inadequate to have this conversation. You need to include physics, cosmology, and religion. Combine this with the fact that philosophy is a DEEPLY insecure discipline that gatekeeps by being deliberately arcane and obscure, and philosophy is just not the right tool to have that conversation.
    1
  4036. Yeah, gonna just slice the list off to the last 20 before even sorting.
    1
  4037. Interesting idea
    1
  4038. THE ENERGY STORAGE PROBLEMS ARE TOO DAM HIGH
    1
  4039. It's really fudged, ngl
    1
  4040. Meaning economy, yes
    1
  4041. With a Cognitive Architecture yes
    1
  4042. Added chapters.
    1
  4043. We can eject into to space with passive radiation via stuff like the SkyCool systems.
    1
  4044. Maybe but there are IP concerns! LOL
    1
  4045. Interesting video but you need to cite your sources.
    1
  4046. Meaning economy
    1
  4047. Between SpaceX and Blue Origin, I think we're only just at the very beginning of the golden age of space exploration
    1
  4048. I definitely believe that about Satya Nadella. Sam is a bit more bullish on AGI and post-scarcity.
    1
  4049. Everyone is building them. I'm showing how to build them safely
    1
  4050. Wash your hands. Seriously.
    1
  4051.  @rays7437  I don't think it matters so long as you use soap
    1
  4052. No, I left it.
    1
  4053. Yeah it is nice to experiment with the system window too
    1
  4054. What did this even mean
    1
  4055. Yeah, we'll see if it goes anywhere
    1
  4056. While that's logical I don't think it's true. Most people have limited cognitive time horizons
    1
  4057. Getting pretty close https://www.perplexity.ai/search/what-s-google-s-latest-math-ai-BHNbQmECRF.24Jt0JEQ_Ew#0
    1
  4058. That's the idea. I'll just keep making videos and communities
    1
  4059. Bro, deflation has already happened. It's why you don't need to work 12 hours a day just to feed yourself.
    1
  4060. Good points
    1
  4061. CoT, RAG, Reflection, MCTS = thats all u need
    1
  4062. So it's programmed to lie....
    1
  4063. JBP is following the footsteps of Seraphim Rose.
    1
  4064. I used the AI to build the AI....
    1
  4065. Well, I mostly play Cyberpunk 2077. But I wake up really early and finish my day early.
    1
  4066. I plan on it. I'm gonna talk with their marketing folks soon.
    1
  4067. That's my point
    1
  4068. That's one approach. I was referring to vicious cycles where each iteration the outcomes become worse such as every time you feed AI data back to an AI system. I will acknowledge I could have said it better, but your definition is not the only definition of a vicious cycle.
    1
  4069. not sure what an MCQ is
    1
  4070. This comment was written by neoliberalism lol.
    1
  4071. My book about autonomous cognitive agents: https://github.com/daveshap/SymphonyOfThought
    1
  4072. go to openai.com and sign up! it's open access :)
    1
  4073. Good question
    1
  4074. No, it's unfortunately not a joke. Google was already losing market share to TikTok https://techcrunch.com/2022/07/12/google-exec-suggests-instagram-and-tiktok-are-eating-into-googles-core-products-search-and-maps/
    1
  4075. It'd be cool to be wrong, but I doubt it
    1
  4076. TEXT-DAVINCI-002 is crazy cheap. This all probably cost less than $0.10
    1
  4077. Lol yes, I don't think anyone argues against the idea of a pause in principle e.g. the optimal solution for people to catch their breath. It's just not possible
    1
  4078. Time will tell
    1
  4079. Please connect on LinkedIn!
    1
  4080. Working on it
    1
  4081. You're talking about decolonization. Yeah, obviously
    1
  4082. I don't think it has much bearing on speed
    1
  4083. Yes, it's the pinnacle of general purpose tech
    1
  4084. fair point
    1
  4085. I have done lots of tutorials and chatbots with this idea. https://github.com/daveshap?tab=repositories&q=&type=&language=&sort=stargazers
    1
  4086. They deliberately exclude knowledge about AI from ChatGPT because "safety"
    1
  4087. This is a great study of narratives.
    1
  4088. What do you mean?
    1
  4089. Like what I need to be admiral?
    1
  4090. Probably not for good
    1
  4091. Thank you!
    1
  4092. Some of that, yes. Automation had a more pronounced effect starting in the 1990s and 2000s in terms of decoupling.
    1
  4093. It should not assume what you do and don't know. It cites its sources so you can delve deeper and ask follow up questions.
    1
  4094. Keep learning English. But this is broadly valuable
    1
  4095. Lol, no
    1
  4096. Look up Teddy Roosevelt
    1
  4097. You already don't have direct power over the government. With AI, you have an opportunity to speak directly with the machine on a daily basis
    1
  4098. Idk
    1
  4099. Industry movements contradict this assertion.
    1
  4100. Had us in the first half
    1
  4101. nope. All definitions of AGI satisfied by September 2024
    1
  4102. Bootleggers and Baptists applies to crypto.
    1
  4103. I left discord.
    1
  4104. Yes, current constitutional AI is limited, but I've done work in it.
    1
  4105. Fortunately this is squarely within fair use
    1
  4106. Consider this: where the hell else is it going to compute right now?
    1
  4107. You know that people said the same thing about coal and steel barons
    1
  4108. Yeah it's really great as a literary tool in hundreds of ways
    1
  4109. Yes.
    1
  4110. I didn't say advancements would stop, just that the breakneck speed of progress will slow down (and is slowing down)
    1
  4111. Equitable does not mean "equality at all costs" like Pol Pot and Lenin.
    1
  4112. I think, biologically, deception emerges in social species. But interesting observation
    1
  4113. One of the few...
    1
  4114. No, actually. I wish I liked Starfield, I really tried. But it's too sandboxy.
    1
  4115. Uh oh
    1
  4116. https://youtu.be/s3I-OxAZXTQ
    1
  4117. The two concepts are orthogonal.
    1
  4118. You don't know that though.
    1
  4119. Test it yourself. No AI has interpreted it that way. Trust objective tests, not your imagination.
    1
  4120. Absolutely zero, there is no incentive to do so
    1
  4121. Gonna have more interviews, but yeah I'm not just your average YouTuber. I was in the trenches
    1
  4122. Yeah, security and trust are critical
    1
  4123. I might give that a try
    1
  4124. Thjat's a great question, I have no idea. You should ask ChatGPT ;)
    1
  4125. Maybe. I found that discovering characters by having them interact with the world, plot, and each other is best
    1
  4126. Theory vs practice. Need experiments
    1
  4127. TRANSFORM AND ROLL OUT!
    1
  4128.  @johnnieblair2325  digital humanities departments are experimenting with this stuff. It will take time to be adopted by the rest of the university culture
    1
  4129. There is a core that is consistent. I have done some experiments to try and elicit different modes or types. But with long enough conversations, there is usually some strong convergence and consistency
    1
  4130. None of them are hiring. They are all firing people right now
    1
  4131. You'll need to inject your data into the prompt
    1
  4132. Dune? Lol.
    1
  4133. My demo GitHub repo: https://github.com/daveshap/qdrant_demo
    1
  4134. I took it as far as I can right now. I circle back to ideas... Sometimes
    1
  4135. The hardest part is finding the most notable papers. We need another layer like a voting system or something.
    1
  4136. I've spoken with some folks at Nvidia. They get it.
    1
  4137. Good point
    1
  4138. Yeah, having finished his book I can say I was disappointed. Good ideas but had too narrow of a lens
    1
  4139. Yes, that's basically the thought experiment
    1
  4140. It's almost certainly just a matter of prompting. This kind of asinine "conclusion" and "gotcha" has happened numerous times but then on closer inspection it turns out the scientists just don't know the first clue about prompt engineering. TLDR I really don't care, I see this kind of nonsense often.
    1
  4141. None
    1
  4142. i mean yeah that's just called AI enabled cybersecurity
    1
  4143. It's weird because most people commenting don't seem to get that...
    1
  4144. China will be held back because all AI must toe the party line
    1
  4145. Yeah but this is like square peg in a round hole
    1
  4146. meted*
    1
  4147. This is the strangest cope I've seen yet.
    1
  4148. There will still be some
    1
  4149. No. It just means there are unseen connections or common undercurrents.
    1
  4150. One day it will be that cool to compose a cognitive architecture. My goal is to create a "plug and play" composable AGI architecture.
    1
  4151. That's making huge assumptions about quantum computing
    1
  4152. No idea, but it is 4x the parameters of GPT-3. It may have a window 10x larger for all I know.
    1
  4153. I've not once heard any definition of AGI that is unlimited. Not sure how you get to unlimited from "human level intelligence"
    1
  4154. The "modernism" definition of marriage is a not-so-thinly veiled attempt to justify the Judeo-Christian assertions about marriage.
    1
  4155. Possibly, we're figuring it out
    1
  4156. Happy to answer questions if you sign up for my Patreon!
    1
  4157. Check out the descriptions for all tiers: https://www.patreon.com/daveshap/membership
    1
  4158. There is now. Wasn't always
    1
  4159. That's why I made this video. I think the world deserves to know as much as possible about an un-elected power-seeking person with this much influence on our future well-being.
    1
  4160. It's an experiment
    1
  4161. I have and plenty of other people have as well
    1
  4162. I actually might change my mind. Was chatting with people on Twitter... undecided
    1
  4163. Even with computational power, training costs are still going up exponentially.
    1
  4164. Thanks!
    1
  4165. Lots of people get a lot of use out of it. I don't personally have any need for it
    1
  4166. No, we gotta push for open source
    1
  4167. Pretty much
    1
  4168. It's almost like you don't understand the point of "explainer" videos on YouTube.
    1
  4169. Yes 🖖
    1
  4170. I'm not worried. I'm on the side of the Western military industrial complex, but I hope to help it transform. I understand the how and why of things as they are, even if they aren't ideal.
    1
  4171. The shortest version is deflation. But yeah I hear you.
    1
  4172. Nothing can simulate that much. This is magical thinking.
    1
  4173. Narratives
    1
  4174. I was focusing on near term
    1
  4175.  @hanziggywuest  just to be safe
    1
  4176. Lots of people have claimed the earth's carrying capacity is 50 billion or larger, but I think they listen to Isaac Arthur a little too closely. From a strictly energetic perspective, sure, the Earth could hypothetically carry a lot more humans. But, uh, energy ain't the whole picture, bro.
    1
  4177. Can't have prosperity without life. ChatGPT already understands that. Go test it yourself
    1
  4178. "It is the mark of an educated mind to be able to entertain an idea without accepting it."
    1
  4179. You get it!
    1
  4180. Embeddings (vectors) are how you search everything. Finetuning is not a good way to learn new stuff. Dreaming, in this case, is more like "consolidating memories during your down time"
    1
  4181. No, those are mostly down to bad prompting and people who don't know how to use the models. I don't really care about small mistakes, people miss the forests for the trees.
    1
  4182. That's why I'm doing what I'm doing. Postnihilism, solving social problems, and contributing to the solution. We have a lot of work to do! Thanks for being here and helping contribute!
    1
  4183. Looks like they'll be around for like 5-10 years more
    1
  4184. 🔥🔥🔥🔥
    1
  4185. Only with anthropic. I'm less and less optimistic about other shops
    1
  4186. Yep, it will work on anything. Someone had the idea of storing each tier of compression so you can zoom in or out.
    1
  4187. HuggingFace INT8: https://huggingface.co/blog/hf-bitsandbytes-integration
    1
  4188. Yeah lots to work on. Which is why I'm super skeptical of any Blockchain or DAO that claims to have figured it out
    1
  4189. Yeah it was Nvidia Broadcast. Fixed it in future episodes
    1
  4190. We have survived everything so far. We have a pretty good track record of figuring stuff out.
    1
  4191. You get the idea lol
    1
  4192. Yeah with a steerable model sure, but you can also finetune custom models to be better at it.
    1
  4193. More than that eventually, but yeah
    1
  4194. No, this is a fundamental misunderstanding. The definition of money has nothing to do with labor. 1. Medium of exchange 2. Unit of account 3. Storehold of wealth I know you mean well but this comment amounts to misinformation. The notion that money is a "claim on labor" or a force of debt is not really accepted by anyone.
    1
  4195. not a lie, people just don't know any better. hence why i made this video
    1
  4196.  @virtly  I do not accept investors or investments of any kind. I have no interest in getting financial entanglements.
    1
  4197. More than you can afford.
    1
  4198. Yes it's pretty simple
    1
  4199. Yes, we need a court jester, contrarian, and skeptical archetypes
    1
  4200. Pretty much yeah
    1
  4201. it happened. you called it.
    1
  4202. Don't make the mistake of assuming your limitations apply to everyone else
    1
  4203. I mean technically, but not everyone wants to just create a youtube or tiktok business
    1
  4204. I'm not finding that post, link?
    1
  4205. Scale AI TransformX Conference: https://scale.com/events/transformx
    1
  4206. Working on that... stay tuned
    1
  4207. Competition will hopefully keep prices down
    1
  4208. There might be diminishing returns though. Limitations in math and coordination. Too much noise to signal.
    1
  4209. That doesn't surprise me, thanks for sharing!
    1
  4210. Instrumental goals emerge based on things like the constraints of engineering and physics. That's the point - it's game theory, not volition. "Instrumental goals" is just another way of saying "Nash Equilibrium". I don't really believe that "terminal goals" are the appropriate way to think about intelligence. I haven't seen any evidence that LLMs are predisposed to think about terminal goals. They will, however, think strategically and tactically, but they are incredibly good at using heuristics, hence my heuristic imperatives.
    1
  4211. I've literally never said that.
    1
  4212. Stonks /s
    1
  4213. No this was all by hand
    1
  4214. Ego is the only thing keeping us human.
    1
  4215. No that is 100% my recorded voice. I did use Audacity to clean it up.
    1
  4216. Eventually. FTL travel may never be possible, which will really change that equation
    1
  4217. You can't "give" an axiom. That's the point. There are fundamental and intrinsic affordances and drives and characteristics about things that will emerge. You FIND axioms, not make them.
    1
  4218. Well what was the result?
    1
  4219. Coffee is spice. Spice must follow
    1
  4220. That won't be an issue soon. Claude has 100k
    1
  4221. No idea what you're talking about.
    1
  4222. Tech progress may continue, but there will be diminishing returns in terms of quality of life. As we approach singularity, we will probably hit a plateau of joy/pleasure/happiness. I could be wrong though.
    1
  4223. Yeah my bad
    1
  4224. The Aspirational layer
    1
  4225. Message me on linkedin
    1
  4226. Some people are saying that, but consider that plenty of people believe the first thing they see on TikTok...
    1
  4227. Time will tell
    1
  4228. Not directly but plenty of financial analysts are undoudtedly using LLMs to read news
    1
  4229. Everyone's entitled to their own opinions. I've personally never found Asimov to be enlightened, entertaining, compelling, or remotely good.
    1
  4230. Nature is way bigger than a theme park. And you already have to wait hours and hours at theme parks.
    1
  4231. Large context window like Claude
    1
  4232. Funny you should mention him...
    1
  4233. Technology has always driven down prices. That's why it "creates jobs"
    1
  4234. Not if we select for a new behavior exclusively from now on. Having children switched from economic necessity to sentimental choice in the last century. That's definitely going to impact evolution
    1
  4235. Thanks for amplifying the signal.
    1
  4236. Are you on 2x?
    1
  4237. Probably before then honestly. Look up "Metahumans"
    1
  4238. Only partly
    1
  4239. Considering there's no global government I'm not sure who you expect to implement a global ROBOT tax.
    1
  4240. ChatGPT is more autistic IMHO lol
    1
  4241. Compassionate understanding doesn't have a mathematical analog, and the difference is semantics
    1
  4242. It's how you fund UBI. You just make everyone a stakeholder.
    1
  4243. Ameca Robot: https://www.engineeredarts.co.uk/robot/ameca/
    1
  4244. Human laws are not morality
    1
  4245. Already did that. Look up ACE Framework
    1
  4246. Yeah but I said it that way because lots of people say the carrying capacity is 50 billion. I agree with you, just addressing the audience where many of them are
    1
  4247. Post Singularity Government? Not a bad idea...
    1
  4248. I'm a bit disappointed. Ten minutes in and neither side has demonstrated a deep systemic understanding of systems. They're both just regurgitating talking points.
    1
  4249. Probably not, those models are too dumb.
    1
  4250. Nah, I don't see humans being economically viable in the long run
    1
  4251. I'm not sure I understand your question. If you just want to talk to GPT-3 without any dressing just use the Playground.
    1
  4252. ah gotcha. yeah you can do it that way. just save all the interactions to log files.
    1
  4253. I do have to point out that Apple is doing just fine right now
    1
  4254. You're welcome. Please consider asking your friends to support me on Patreon so I can do this full time https://www.patreon.com/daveshap
    1
  4255. Most humans are unable to discover new knowledge, though. They lack the training, intelligence, and willpower.
    1
  4256. I'll try a larger camera window.
    1
  4257. It probably won't
    1
  4258. Yeah, trust is critical. Eventually though, especially as children grow up with robots, trust in machines could go up
    1
  4259. Don't tell me what to do 🤪
    1
  4260. Yes, but what is the primary thing that politics argued about?
    1
  4261. Glad it helped!
    1
  4262. Back again
    1
  4263. You would probably fit in well at my research discord. Lots of people to help out: https://discord.gg/5VNRzthf
    1
  4264. I want to retire by 42
    1
  4265. Hope you got a warranty
    1
  4266. Well...
    1
  4267. Children can't inherit businesses until they're adults, at which point they are legally independent.
    1
  4268. Yes, we're heading for a desk-flip moment
    1
  4269. Not familiar with it
    1
  4270. B2B
    1
  4271. FOMO and hype cycles. A single day of change does not make a business cycle
    1
  4272. GitHub for this experiment: https://github.com/daveshap/AlignmentAgentModel
    1
  4273. that would kill me
    1
  4274. That's the goal
    1
  4275. Read Life 3.0
    1
  4276. Wow! Thank you!
    1
  4277. this comment will age poorly lol
    1
  4278. I dunno man. I'm a paid subscriber to ChatGPT and have even been using the API and ChatGPT seems to be getting dumber and more useless. Claude is now easily 10x less effortful to use.
    1
  4279. You have to treat other models similar to vanilla GPT-3, they are basically just autocomplete engines
    1
  4280. 2025 or so, yeah
    1
  4281. Depends on how hot it gets. We've seen very short term engagements
    1
  4282. Yeah that's why I didn't lead with this, and characterized that Darwinism in AI will have very different pressures
    1
  4283. I tend to agree. That's why I was like "All this is over and done with by 20 years"
    1
  4284. This one? https://github.com/kyb3r/emergent
    1
  4285. Lol, lots of people valorize physique dude
    1
  4286. I was hoping no one would notice...
    1
  4287. Agreed, but my videos about singularity stuff don't get as much traction. The world isn't ready ;)
    1
  4288. Thanks man! That's why I am here! Just want to help
    1
  4289. So is Nancy Pelosi's entire retinue
    1
  4290. Original Paper: https://onlinelibrary.wiley.com/doi/10.1002/aaai.12064
    1
  4291. Haven't read it. What do you think about it?
    1
  4292. Yes, based on AMQP
    1
  4293. Some alignment is built into the original data (humans are far more benevolent than malevolent) so just on raw data it's going to be biased towards goodness and helpfulness.
    1
  4294.  @alan2here  oh! That is because this is all the INSTRUCT series. They are fine-tuned to follow instructions
    1
  4295. What do you mean by "typical decoding"?
    1
  4296. It described what it wanted.
    1
  4297. I've talked about some of these points several times yes
    1
  4298. heeeeeeeyoooo
    1
  4299. It's going to take off fast
    1
  4300. Fixing punctuation with GPT-2
    1
  4301. I do have a discord but it's pretty much silent. Maybe I'll clean it up and see if anyone wants to join
    1
  4302. https://discord.gg/creduKKK let's see how this goes
    1
  4303. Moving the goalposts
    1
  4304. OODA loop = TRC (Terminal Race Condition). Thanks for the good video.
    1
  4305. English
    1
  4306. You're making a heck of a lot of assumptions about what people want.
    1
  4307. U DONT SAY :D
    1
  4308. Optimistic nihilism is, unfortunately, a pipe dream.
    1
  4309. Yep. Just a bit of information injection makes it hugely more useful
    1
  4310. Actuarial Solipsistic Intelligence
    1
  4311. My book Symphony of Thought: https://github.com/daveshap/SymphonyOfThought
    1
  4312. I use them sometimes but the purpose of my videos is to teach
    1
  4313. If CODEX or COPILOT had a plugin for NPP I would give it a try
    1
  4314. I didn't say catastrophic
    1
  4315. What do you mean?
    1
  4316. Others are working on it
    1
  4317. I've got quite a few videos in my backlog plus a few books, but CA videos don't get as many views.
    1
  4318. Not zero, but lower. Materials will get somewhat cheaper due to energy and robotics, but they are still heavy and valuable.
    1
  4319. I know...
    1
  4320. I'm pretty sure this is why my wife married me. She wants a tachi
    1
  4321. yeah workshopping a better term. I think we're settling on "volitional"
    1
  4322. Interesting concept. I would think that understanding capabilities is crucial to understanding alignment.
    1
  4323. That all makes sense, thanks for sharing (as always)
    1
  4324. Glad it helps! Feel free to shoot over any questions you have, I am always looks for more ideas and problems to solve
    1
  4325. Eastern Europe is second world AFAIK. I was thinking more Africa and rural India.
    1
  4326. Probably
    1
  4327. NOOOOOOO
    1
  4328. Orthogonality. The resources we consume will eventually be trivial to them. Just like how you don't worry about the ants eating your lunch.
    1
  4329. You're missing the point. Burnout is recognized as an HPA axis disorder. Same thing.
    1
  4330. This is the Way
    1
  4331. I don't think I agree with your initial characterization about sentience and consciousness, but the rest of your comment is pretty good.
    1
  4332. Five figures sounds right. Just make sure they've actually done the kind of work before.
    1
  4333. If I were at Google right now I would be seriously considering heading for the exit.
    1
  4334. https://www.hollywoodreporter.com/business/business-news/authors-sue-meta-openai-class-action-1235588711/
    1
  4335. Epistemic grounding
    1
  4336. Yep. Gotta fund it without causing inflation
    1
  4337. Evolution is a really strange differentiation. Why would that make a difference?
    1
  4338. They are coming
    1
  4339. He posted just before me but I watched his right after I posted. We came to very similar ideas
    1
  4340. Agree to disagree. I think that sentience (and consciousness) are scalar, composed of multiple dimensions and come in degrees.
    1
  4341. Are you using long chats?
    1
  4342. My goal is still to delete my channel one day, but the work isn't done
    1