Comments by "clray123" (@clray123) on "Lex Clips" channel.

  1. Do not exaggerate, these days you can finetune an average AI in a matter of a few hours.
    178
  2. This applies to all CEOs. In terms of honest opinion you could just as well talk to a LM trained on corporate marketing material.
    29
  3. Such deep answers... they seem to have trained ZuckAI using collected outputs from GatesAI.
    18
  4. The LLMs are to programmers as automatic checkout lanes in supermarkets are to cashiers. Yes, you can check out through the machine, but most people still prefer to be served by a cashier because (1) it requires less effort and (2) when anything should go wrong it's so much less pain in the ass. For the same reason, good companies will retain their good programmers while shitty companies will generate "something" with LLM (just like they employ mediocre programmers today).
    9
  5.  @chrismacleod9326  If you had ever reviewed someone else's code, you'd know that having the code generated is not a "big reduction" of effort at all. Understanding and proof-reading foreign code is in fact in most instances way more difficult than just writing your own.
    6
  6.  @barstar888  You're wrong, no Apple products will ever be in my home. Nor any Facebook products, of course.
    6
  7. Aliens are manifesting themselves during a DMT trip in the same sense as the good old JC manifests himself on a molding piece of bread.
    4
  8. When a 170 billion parameter AI is telling you that a blade of grass has one eye, you can be pretty sure it's not reasoning.
    4
  9.  @jamcowl  Maybe his friends had short positions at that time.
    3
  10. The unsettling thing is that it's fkin around with systems we don't fully understand, and the consequence of getting it wrong is people getting sick or dying. It's a bit as if a software engineer started changing some important production code here and there to "maybe" fix some problem, with no backup or rollback plan whatsoever.
    3
  11. @Raw Engineer For example, imagine a virus which does not really hurt people with Asian genome all that much, but wreaks havoc in other populations...
    3
  12. If you can easily trick something into inconsistency and contradictions and factual nonsense, you can be pretty sure it's not reasoning. It's just a clever playback device.
    3
  13.  @grassandglobs  The computer-generated texts are nonsensical on a much deeper level, revealing a real lack of awareness of the topic and involved entities (such as changing pronouns used to refer to the same entity in two subsequent sentences).
    3
  14. @Language and Programming Channel It does not learn any ideas, it just learns which words/phrases are likely to appear to together and then replays those when you prod it using other words. Sometimes it just replays whatever it has recorded without even changing it a bit.
    3
  15. Uhh you are aware than in an interview like this the questions that are going to be asked are made known to the interviewee beforehand?
    2
  16.  @slickmeleon69  The fact that it is an excerpt from a larger conversation does not contradict at all what I said. When you invite an important talking head to a conversation he gets a list of topics beforehand otherwise he won't accept your invitation to start with. And if you ask surprise questions and God forbid do not allow the marketing team to cut out the responses that they don't like afterwards, then it will be the last time anyone from the company gave any sort of interview to you.
    2
  17.  @joebob6953  What if pigs could fly? Would you get on board an aircraft which contains some AI-generated code, given that the AI "just sometimes, very rarely" makes silly mistakes like claiming 2+2=5? This is where we stand today. And the most funny part is that the designers of these AI systems don't really know WHY these silly mistakes are made. At best they cover them up when they are found. It's "screwed by design", so to say, (un)provably wrong. But yes, it's gonna replace some trivial applications in which correctness does not matter. There's a good reason why AI content generation it is replacing various "artists" first (because unlike with an aircraft nobody really cares if an artist's paint is slightly off-hue).
    2
  18. So, I debugged a problem today, took about an hour of work. One system reported repeated errors always around 2am at night - an attempt to authenticate via REST timed out. The reason turned out to be a backup process locking the database of the invoked system during the same period. The invoked system executed DROP TABLE instead of DROP TEMPORARY TABLE (the latter would have avoided the lock). But it only started causing problems after a recent upgrade of the database to MySQL 8 and related changes to online backup . The fix was to modify a few lines in the buggy SQL generation module which should have (always, even before the bug) generated DROP TEMPORARY TABLE but has not. The system that was changed has some 200K lines of code overall. The analysis involved two people checking logs in both the invoking and the invoked system and one person (me) enabling even more logging and performing targeted tests in development environment to confirm the hypothesized cause-effect, along with reading database documentation and remembering that the problems began to occur after db upgrade (despite false information about it from the affected system, which only started reporting errors about a month later). This sort of thing is ROUTINE work for devops. Not generating of some shitty half-working code snippets. Until I see an AI capable of performing reasoning and communication of the kind described above, I will not lose any sleep about the safety of my job.
    2
  19. I guess the idea is that in future ChatGPT will interview clients to help them figure out what they want. And then implement it.
    2
  20.  @OnEiNsAnEmOtHeRfUcKa  nah, AI just rolled the dice... and that's all it does, rolling dice and constraining results to make them appear not completely random
    2
  21. Except because biological systems are truly perplexingly complicated with lots of implicit dependencies you don't know the consequences of your "nano-precise" actions. (Well, you do know the immediate genomic consequences full well, but you don't know wtf side effects are going to happen in the long term after you do these actions.)
    2
  22.  @fast_harmonic_psychedelic  That's not true, we know very well what happens when genes combine naturally because we have been observing the outcomes for thousands of years. Same goes for uttering strange words in front of people. However, randomly messing around with a complex system we don't understand - but of which we do understand enough to infer that our actions WILL perturb the system in significant ways, that's just asking for trouble. The correct comparison would be to messing around with random code lines in a huge software system - there too you wouldn't know the exact consequences, but you could be sure, based on what we know about software, that changes would not be inconsequential and also pretty sure that consequences of arbitrary changes would be much more likely bad (breaking something somewhere) rather than good by accident.
    2
  23.  @fast_harmonic_psychedelic  BTW, the doctors at Unit 731 also acted in the name of scientific progress. The more fitting term however is crimes against humanity, and hopefully the perpetrators will always be caught and punished.
    2
  24.  @fast_harmonic_psychedelic  The doctors very much believed they were doing high grade science. In fact some of them were imported after war and continued their "research". The point here is that "scientific progress" is a weak excuse for engaging in activities with high potential of causing human misery and that it's never either ethics or science, you need both.
    2
  25. @Raw Engineer I wouldn't be so sure about that - if anything, a totalitarian state has a bigger risk of revolt than a democracy - that's why it needs to remain totalitarian in the first place. So the virus would indeed need to be carefully targeted so as to allow declaration of "victory" rather than collapse of authority.
    2
  26. Is all cognition/reasoning like that? Quite obviously it's not, given that very few humans have read "the whole Internet" and we can still imagine things and put together reasonable sentences fairly easily.
    1
  27. Well, he's a failed physicist for one, most probably holding some big personal grudges against many people and post hoc rationalizing his failure.
    1
  28. Pretty much so, and when you have everything in material wealth and political power, the only thing you aren't allowed as an old rich prick is depravity with young people (ok, the other missing thing is ability to be young again and enjoy great physical health, but this is something that no money can currently buy). So it's somewhat expected that they gravitate toward such forbidden fruit, especially if they have the sort of mentality that they personally "deserve" to get whatever they want (which doubtlessly helped propel them into the high places of "success" to begin with). I'd say it's not so much about their wish to be evil, but simply a continuation of elevating themselves above other people, regardless of whether it causes suffering to others (possibly they would also discount that suffering, e.g. give the victim money or gifts as a compensation). Paradoxically, the definitions of taboo behaviors and rules that forbid them are what allows such behaviors to be both counted as a very special prize, a symbol of trust among the chosen, and a subject for blackmail. What I find pathetic is that putting your old dick in a kidnapped and drugged teenager is supposed to be worth all this effort and risk simply because everyone says "you can't do it". I guess my hormones are just not strong enough to understand the attraction.
    1
  29.  @JamieD1298  It doesn't even have to be confirmation bias in the sense somebody told you so. It can be confirmation bias on a basic biological sensory level, just like our circuits are superefficient in recognizing faces, they can be superprone to confabulating figures/entities. What I'm saying is that this is not a glimpse of some "external reality" or "higher beings" sending signals to us. Matthew also gave a solid argument to the same effect - no new secrets are ever revealed during the trips that could be validated after it's done. That subjects hallucinate in similar ways reveals something about the subjects, but nothing more. I'm very surprised that Lex is even entertaining such bollocks, but maybe he's doing it for the sake of being entertaining (I hope so).
    1
  30. Extreme embarrassment and shame, according to the dictionary.
    1
  31.  @ThomasTomiczek  It rather sounds like it is you who is talking bullshit. Cognitive loop? These things are currently tripping up on their own randomly generated bs, with devs desperately trying to mask it by glueing not some very intelligent AI but-rather clunky old traditional hardcoded logic workarounds on top of it to hide the embarrassment. Also slow and memory inefficient as hell, compared to handcrafted algorithms, say, for parsing or compiling good ole machine languages.
    1
  32. They improved his alignment using RLHF.
    1
  33. Business people/academics talking about stuff they really have no great practical experience with while borderline insulting people who do and who they depend on. That is surely the way forward to a great programmer-free future... just make sure you don't accidentally insult too many programmers in your own company or else you might only be left ChatGPT's virtual shoulder to cry on.
    1
  34. Natural language is inherently crap at communicating ideas clearly. I want to see the half-imbecile business people actually talk to the AI to get their requirements across, will be great fun to watch.
    1
  35. It's just too bad that to believe in conspiracies, for the most part, is idiocy as well.
    1
  36. Pls next time run some perplexity metrics on this model before publishing its responses.
    1
  37. You don't really need lucifers to explain fucked up brains, a MRI scan is probably enough these days to spot psychopathy.
    1
  38.  @derekshirley5543  I have no idea about you mean by "woke culture", this is the first time I heard the term. I'm just saying that you don't need any metaphysical mumbo-jumbo for explaining something which can be just as well (probably better) explained and predicted by our understanding of nature.
    1
  39.  @derekshirley5543  Yes, I'm also noticing the new overarching (presumably automated) online censorship, and I'm certainly against it. As for metaphysical/spiritual, I see no use of such antiquated concepts when better explanations are available. At best they are a vague description of what we can observe, but they predict nothing actionable, so they don't really contribute much value. To be more concrete, if we know that a person has a psychopathic brain and is incapable of empathy, we may wish to keep a close watch on the so afflicted person. But if all we can say is that "it's (the d)evil", we can't do anything because there no reliable test exists for "devilhood" (and in fact all the unreliable "tests" we historically invented only contributed more to humanity's misery).
    1
  40.  @derekshirley5543  In fact brain research and psychology can explain and even predict the atrocities and all the associated human behaviors (e.g. complacence in face of authority, various reactions under great duress) rather well. The same can't be said about stormy angels, though.
    1
  41.  @derekshirley5543  Nobody is presuming such a thing.
    1
  42.  @derekshirley5543  Ok, you can believe whichever fairy tale brings consolation to you. If your mathematical "God" is real, then "he" doesn't care about things such as "good" and "evil", as is quite clear from study of nature that "he" "created".
    1
  43.  @derekshirley5543  Yeah right, fear. That's what the book tells you. So you can fear all you want. Good bye.
    1
  44.  @derekshirley5543  Fear is only appropriate where there is a danger of harm. OTOH, irrational fear of imagined things is not appropriate, and most people do grow out of it in late childhood. Unless of course they are regularly brainwashed to sustain such fears.
    1
  45.  @derekshirley5543  There is no soul. There is no hell. The only danger is your body suffering while you're still alive.
    1
  46.  @derekshirley5543  I will also add that both good and evil are man-made concepts, used to describe what we like and don't, based on the aforementioned perceived danger to our bodies. It's really that simple and no holy books required to grasp.
    1
  47.  @derekshirley5543  Huh? I just explained to you why rape, murder, or even psychological damage are "evil". Maybe you should take your pills, though, you seem "a little" unhinged.
    1
  48. When you mess around with an artificial neural net, it produces messed up "confabulated" results. When you mess with your brain, it produces essentially the same kind of results as a messed up neural net would, just more so because it's bigger/more complicated. I'm sure it's a huge experience for the experiencing individual, but that's about how "profound" it is.
    1
  49.  @armin3057  I'm pretty sure Elon Musk does care when it can actually hurt his pocket... if he would not, he would not be a billionaire. There's very good chances his "unhindered" twittering is very much premeditated so as to make money off these "slip ups". Like all/most CEOs the man is a manipulator, but he prefers hiding behind an easy-going, down-to-earth impulsive dude persona (glibness and charisma being well-known sociopathic traits).
    1
  50.  @SafetyLucas  What you say is true under assumption that sacrifice is what is expected of people. But who is there to tell you how much you should sacrifice? How about a non-sacrificing life of a happy simpleton, is it a bad life, and if so, who decides that?
    1