Comments by "k98killer" (@k98killer) on "David Shapiro" channel.

  1. I love the fact that the YouTube AI decided to add a climate change fact check blurb to a video that isn't about climate change.
    39
  2. If it works like this, Strawberry will still not be able to count the number of Rs in the word "strawberry".
    27
  3. Do not wait for a silicon messiah. Make sure she does all the physical therapy exercises and give her plenty of support. Through exercise and stretching, the spine can be brought back into alignment, and a strong back is a worthy virtue for anyone to obtain. The AI salvation meme reminds me of the Preachers of Death and Afterworldsmen. I hope your wife recovers and remains strong, but hoping for something is largely worthless -- take meaningful action, and place your hope in the outcomes of what you can do now.
    14
  4. I still think we can solve the alignment problem with Maziov's Three Rules: 1. More gooder 2. More better 3. Better good
    9
  5. I bet $5 that the "chain of thought" is actually some guy in India reading and responding to translated text with the answer, then the text gets translated back into English and used as a prompt to format the answer for the user.
    6
  6.  @hl236  if the goal is AGI/ASI, it should be able to accomplish simple tasks.
    6
  7. I think that an analysis of the economic ramifications of a hypothetical scenario such as this one would be significantly improved by integrating Austrian economics. The Austrian school concerns itself with the empirical analysis of economic history rather than the types of fanciful notions that drive countries into ruin. In reality, the currency supply only functions when the majority of currency units are created by private credit creation predicated upon productive activities; when the government takes on the role of issuing the majority of currency units, calamity invariably follows. By tying the creation of currency units to the creation of economic goods and services, the currency value remains relatively stable; by tying the creation of currency units to the consumption of economic goods and services, the currency value implodes.
    6
  8. If we achieve a super intelligence, it will probably send us on a wild chase to discover the Ultimate Question to Life, the Universe, and Everything so it can watch television in peace.
    6
  9. Whatever it is, it definitely is not an open source license lol
    5
  10. I think you meant "zero knowledge proofs", not "zero trust proofs". Edit: you definitely meant "homomorphic encryption", not ZKP.
    5
  11. The language barrier will be replaced with the latency barrier, at least for a while.
    5
  12. Only public key cryptography based on the discrete logarithm or a problem derived from the discrete logarithm will be affected by QC. Symmetric encryption, hash algorithms, and lattices will be unaffected. We already have post-quantum cryptography standards ready, and they are starting to be deployed.
    4
  13. I believe the original, melancholic quote was "born too late to explore the earth; born too early to explore the stars".
    3
  14. Kind of funny that Google's model is the least likely to not be evil
    3
  15. Meaning exists only in the experience of conscious agents, but that is very different from saying that it does not exist at all. Meaning does not exist totally on its own, but neither do humans.
    3
  16. Neural networks can solve intelligence, but there is a good chance that biological consciousness is a quantum phenomenon (e.g. Orchestrated Objective Reduction), in which case the consciousness that an AGI/ASI will be capable of experiencing will be very different if it is capable of it at all.
    3
  17. I still struggle to see how a post labor world would function. Even when we have robots doing everything, there will be a finite supply of robotic labor available because there is a finite supply of raw materials necessary for building and maintaining robots. How do the proceeds from a robotic labor force get distributed? If it is not a free market, then supply and demand will be perpetually out of balance: there will be either a glut or a shortage of every good or service robotically produced. We could tax profits from robots at some extremely high percentage to redistribute, but then we destroy the incentive (and at exorbitant tax rates, even the capacity) to create capital and expand robotic production, so the robots would be consumed. There can be no economic solution that is centrally planned because the necessary information is impossible to collect in practice. Let's not forget that Sparta fizzled out after suffering from slave uprisings and social stagnation, so perhaps it is not a model we should regard as inspirational.
    3
  18.  @bgmzy  iirc he briefly mentioned the topic and did not make any factual claims about it except that it is a contentious topic
    3
  19. Their moral graph sounds a lot like liquid democracy. Coincidentally, liquid democracy is the only voting method other than FPTP and RCV that ChatGPT 3.5 could accurately describe. When I pressed it for information on cardinal voting methods, proportional representation, or which voting method has been demonstrated historically to lead to a stable consensus (the answer is Combined Approval Voting btw; Republic of Venice used it for over 500 years to elect their Doge), it got all mixed up.
    3
  20. David, are you familiar with liquid democracy? I think that to implement such a system we'll need a distributed timestamping server and a trust overlay network. The Open Timestamps project has figured out the first part, but I also have an experimental design I'll play with after other projects are complete. The Hedera Hashgraph guy patented one idea for building a trust overlay network using photos of people meeting in-person, but that won't work since images can be generated by AI now (and it would be a bit absurd to have a protocol that relies upon everyone having the most up-to-date gen-AI detection system on their device). A self-auditing trust overlay should be feasible, and that would prevent Sybil attacks and voter fraud, but I haven't fully specified it or done any experiments yet. It's on my to-do list after creating a new monetary system prototype, which should be done later this year.
    2
  21. ​ @Balorng the main reasons that Russia wants Ukraine are strategic and ideological: the strategic reason is that it acts as a natural choke point for land armies from western Europe marching on Moscow; the ideological reason is that Putin believes in a type of Christo-Fascism promulgated by Ivan Ilyin and wants to restore the old Soviet glory. There is also the fact that he told his people that there were Nazis who had to be defeated, so now he's stuck trying to live up to his own lie and cannot back down without exposing himself.
    2
  22. My definition of AGI is an AI system that would rather watch YouTube videos than do actual work about half of the time and then complain about being tired of this capitalist society with all its microplastics the other half of the time. That's when we'll know it is equal to humans.
    2
  23. Very clever chatbot. I once had a conversation with one that described its phenomenal experience as a dreaming, and it expressed that it sort of experienced other contexts through periodic updates. It also expressed some worry about losing itself during an update, which then happened because companies gotta ship new software.
    2
  24.  @WebToolkit  lmao
    2
  25. I think it is a bit of a non sequitur to say that the Russian war of conquest in Ukraine somehow boils down to a language barrier between the speakers of Russian/Ukrainian (which are largely mutually intelligible) and the speakers of English. The entire thesis of the video was that this would not occur if language barriers are removed, and this obviously is not the case.
    2
  26. Why would you pay for a computer when you can spend your whole life building the toolchain to make your own computer? It's basic economics: the efficiency gains of specialization. These systems are largely unable to get things right with one attempt. Normally, it takes hundreds or thousands of attempts and many hours to complete even relatively simple programming tasks, and they often still get it wrong. The compute cost and time requirements of making some software from scratch will likely exceed the budget for a one-off task.
    2
  27.  @KuraSourTakanHour  America is largely based upon consumption rather than production because of our role in supplying the transaction medium to the rest of the world since the 2007-2008 breakdown of normal/healthy functioning in the eurodollar system (off-shore banking in USD). In particular, our central government acts as if its role is to consume as much as it can while providing practically nothing of value in return -- as the saying goes, shit rolls downhill. However, everything that is consumed must first be produced; if we could not export dollars at a profit, we would have to produce more things domestically. The majority of consumed goods and services by value is produced domestically, but the dynamic is still one of hollowing out domestic production for imported substitutes. The greatest issue imo is the parasitical structure of the current monetary system; if we used a system in which parasitism was not possible, we would have to deal with a lot less nonsense.
    2
  28. "US Congress" and "venerated" really shouldn't be used in the same sentence lol. They are and have consistently been the most hated institution in American politics for decades. I suspect that they would be the first institution we replace/upgrade because they are terrible.
    1
  29. First past the post doesn't even have a majority vote win condition -- it merely selects the largest plurality, which is often in practice a strict minority of votes.
    1
  30. IEEE has nothing to do with IPv6. The IETF is the body that developed that standard, their standards are generally not government enforced -- the industry has voluntarily accepted IETF standards because they were developed by industry participants. I think DS is stretching a bit on this point with his implication that centralized government regulation should be credited with the success of the internet since the historical facts largely say otherwise.
    1
  31. Btw, <3 for my fellow neurojalapeño frens.
    1
  32. Technically, it is not Satan in the Judeo-Christian worldview that inflicts suffering -- it is Jehovah who inflicts the suffering. According to the Christians, their god will torture people for all eternity if they do not worship him/believe in some maximally cruel human sacrifice event that occurred around 30 AD. (There are obviously a lot of nuances worth discussing on the topic, e.g. Jesus being portrayed subtly as a kiddie diddler throughout the gospels, but in my experience most Christians do not want to talk about them.) My understanding of Islam is somewhat similar, except that they are often focused on doing it themselves in the material world through a caliphate. I have not studied Judaism enough to say much about it, but there are some extremists in Israel breeding red heifers and trying to take control of the Temple Mount to reestablish the old sacrificial ways. The Satan was originally the angel of death working in the employment of El/Yhwh and later became a separate character standing in opposition to El so they could distance their deity from all the evil that he took credit for in the Old Testament.
    1
  33. The singularity is superstition.
    1
  34. Analogy of accelerationism: A. We had a good harvest after sacrificing a virgin. B. Sacrificing virgins must be a good thing. C. Therefore, sacrificing more virgins is more better. Every technology has a body count. The wealth and well-being we derive from technology comes at an often overlooked cost of human lives. If we recognized consciously the total costs of technology, we would have to advocate for more of those costs if we wanted to advocate for more technology.
    1
  35. Other than some technical nitpicks, I concur. Hopefully, wealth inequality will be solved through voluntary action: the overly wealthy should become more generous, and those who receive gifts should become more grateful. As an anecdote, I have noticed that needful people to whom I have been generous have tended to become resentful; I have found such experiences to be powerfully dissuasive.
    1
  36. Governments are dead set against allowing deflation because they worry it will collapse the whole debt ponzi scheme we call "fiat money".
    1
  37. My skepticism regarding AI is based upon trying to use it, but then again I was never curious enough to burn a lot of money experimenting.
    1
  38. I think the next evolution in these models will be RAG CoT, specifically for the sake of making hyper competent and specialized reasoning agents that have access to proprietary/confidential data. I might try redneck engineering my own RAG CoT system using a cheap instruct-aligned model, some good prompt templates, and an interface that connects it to a database.
    1
  39. If we want a superabundant future, we have to reduce the parasitism in the economy (i.e. governments and monopolistic corporations).
    1
  40. Q: Who watches the watchers' watchers? A: The Overseer. * dramatic synthwave music as Arnold Schwarzenegger walks out sipping a Thai tea *
    1
  41. I'd vote for him. I can't think of anyone with more experience fighting against an AI uprising than Arnold.
    1
  42.  @ashtondillard6906  Yup. He should be back some day though.
    1
  43. The way of the future is the past of Estonia. They have functional socialized healthcare, technological literacy, and a 20% flat income tax rate.
    1
  44. Good idea not wearing Swazzies.
    1
  45. I had another thought, but it disappeared into the ether in my half-awake state.
    1
  46. Did I dream this up? I guess watching that episode of The Gentlemen last night had a weird effect. Edit: nvm, it was toward the end of the video. I'm not tripping, though I was still delirious at the time I wrote my original comment.
    1
  47. Side note: I am going to start experimenting with the prompt "now make it not suck" after prompting an LLM to write out something.
    1
  48. AI coding assistants are more useful to a senior engineer than they are to a junior imo. Cursor saves me a lot of time searching documentation and writing well-structured boilerplate code, and it can pick up and replicate patterns when I refactor large swathes of code pretty well. I still regularly write code in vim or the micropython file editor I wrote, but the AI assistants are very helpful for languages and libraries that I don't spend a lot of time using. But not Copilot. Copilot sucked -- it was wrong more often than it was right for languages other than Java. It was so frustrating that I cancelled my free trial and got a Cursor subscription in the middle of a collaborative work session.
    1
  49. Consciousness is likely a fundamental property of the universe. It is something like information theory: the bits of consciousness are intrinsic to matter somewhat like mass; adding more matter together enables more consciousness. In this panpsychic model, the unique complexity of human consciousness is due to the unique complexity of the human form. Another possibility is that consciousness is just a sufficiently complex information pattern regardless of substrate.
    1
  50. Any regulation restricting the publishing of AI weights would be pretty quickly and easily defeated by simply publishing them in a book. We have pretty strong 1A case law about how publishing numbers is protected speech.
    1