3 ms·

“It’s just math underneath” This quote is so telling. I’m going to be straight with you and this is my opinion so you’re free to disagree. From my POV you are

“It’s just math underneath” This quote is so telling. I’m going to be straight with you and this is my opinion so you’re free to disagree. From my POV you are out of touch with the ground truth reality of AI and that’s ok because it has all changed so fast. Everything in the universe is math based and in theory even your brain can be fully modelled by mathematics… it’s a pointless quote. The ground truth reality is that nobody and I mean nobody understands how LLMs work. This isn’t me making shit up, if you know transformers, if you know the industry and if you even listened to the people behind the technology who make these things… they all say we don’t know how AI works. But we do know some things. We know it’s not a stochastic parrot because in addition to the failures we’ve seen plenty of successes to extremely complicated problems that are too non trivial for anything other than an actual intelligence to solve. In the coming years reality will change so much that your opinion will flip. You might be so stubborn as to continue calling it a stochastic parrot but by then it will just be lip service. Your current reaction is normal given the paradigm shift happened so fast and so recently.

mjr00 11mo ago
> The ground truth reality is that nobody and I mean nobody understands how LLMs work. This is a really insane and untrue quote. I would, ironically, ask an LLM to explain how LLMs work. It's really not as complicated as it seems.
- rybosworld 11mo ago
  It's not an insane thing to say. You can boil LLM's down to "next token predictor". But that's like boiling down the human brain to "synapses firing". The point that OP is making I think, is that we don't understand how "next token prediction" leads to more emergent complexity.
  - mjr00 11mo ago
    The only thing we don't fully understand is how the ELIZA effect[0] has been known for 60 years yet people keep falling for it. [0] https://en.wikipedia.org/wiki/ELIZA_effect https://en.wikipedia.org/wiki/ELIZA_effect
    - rybosworld 11mo ago
      > The only thing we don't fully understand is It seems clear you don't want to have a good faith discussion. It's you claiming that we understand how LLM's work, while the researchers who built them say that we ultimately don't.
- threethirtytwo 11mo ago
  https://futurism.com/anthropic-ceo-admits-ai-ignorance https://futurism.com/anthropic-ceo-admits-ai-ignorance There’s tons more where that came from. Like I said lots of people are out of touch because the landscape is changing so fast. What is baffling to me is that not only are you unaware of what I’m saying but you also think what I’m saying is batshit insane despite the fact that people in the center of it all who are creating these things SAY the same thing. Maybe it’s just terminology…understanding how t build an LLM is not the same as understanding why it works or how it works. Either way I can literally provide tons and tons more of evidence to the contrary if you’re still not getting it: We do not understand how LLMs work. Also you can prompt an LLM about whether or not we understand LLMs it should tell the same thing I’m saying along with explaining transformers to you.
  - mjr00 11mo ago
    That's a CEO of an AI company saying his product is really superintelligent and dangerous and nobody knows how it works and if you don't invest you're going to be left behind. That's a marketing piece, if you weren't aware. Just because the restaurant says "World's Best Burgers" on its logo doesn't make it true.
    - threethirtytwo 11mo ago
      Didn’t I say I have tons of evidence? Here’s another: https://youtube.com/shorts/zKM-msksXq0?si=bVethH1vAneCq28v https://youtube.com/shorts/zKM-msksXq0?si=bVethH1vAneCq28v Geoffrey Hinton father of AI who quit his job at Google to warn people about AI. What’s his motivation? Altruism. Man it’s not even about people saying things. If you knew how transformers and LLMs work you would know even for the most basic model we do not understand how they work.
      - mjr00 11mo ago
        I mean at a minimum I understand how they work, even if you don't. So the claim that "nobody and I mean nobody understands how LLMs work" is verifiably false.
fingerlocks 11mo ago
Try to use an LLM to a solve a novel problem or within a domain that can’t easily be googled. It will just spew over-confident sycophantic vomit. There is no attempt to reason. It’s all worthless nonsense. It’s a fancy regurgitation machine and that will go completely off the rails when it steps outside of it’s training area. That’s it.
- threethirtytwo 11mo ago
  I’ve seen it solve a complex domain specific problem and build a basis of code in 10 minutes what took a year for a human to do. And it did it better. I’ve also seen it fuck up in the same way you describe. So do I weigh and balance these two pieces of contrasting evidence to form a logical conclusion? Or do I pick and choose one of pieces of evidence that is convenient to my world view? What should I do? Actually why don’t you tell me what you ended up doing?

mjr00 11mo ago

> The ground truth reality is that nobody and I mean nobody understands how LLMs work. This is a really insane and untrue quote. I would, ironically, ask an LLM to explain how LLMs work. It's really not as complicated as it seems.

rybosworld 11mo ago
It's not an insane thing to say. You can boil LLM's down to "next token predictor". But that's like boiling down the human brain to "synapses firing". The point that OP is making I think, is that we don't understand how "next token prediction" leads to more emergent complexity.
- mjr00 11mo ago
  The only thing we don't fully understand is how the ELIZA effect[0] has been known for 60 years yet people keep falling for it. [0] https://en.wikipedia.org/wiki/ELIZA_effect https://en.wikipedia.org/wiki/ELIZA_effect
  - rybosworld 11mo ago
    > The only thing we don't fully understand is It seems clear you don't want to have a good faith discussion. It's you claiming that we understand how LLM's work, while the researchers who built them say that we ultimately don't.
threethirtytwo 11mo ago
https://futurism.com/anthropic-ceo-admits-ai-ignorance https://futurism.com/anthropic-ceo-admits-ai-ignorance There’s tons more where that came from. Like I said lots of people are out of touch because the landscape is changing so fast. What is baffling to me is that not only are you unaware of what I’m saying but you also think what I’m saying is batshit insane despite the fact that people in the center of it all who are creating these things SAY the same thing. Maybe it’s just terminology…understanding how t build an LLM is not the same as understanding why it works or how it works. Either way I can literally provide tons and tons more of evidence to the contrary if you’re still not getting it: We do not understand how LLMs work. Also you can prompt an LLM about whether or not we understand LLMs it should tell the same thing I’m saying along with explaining transformers to you.
- mjr00 11mo ago
  That's a CEO of an AI company saying his product is really superintelligent and dangerous and nobody knows how it works and if you don't invest you're going to be left behind. That's a marketing piece, if you weren't aware. Just because the restaurant says "World's Best Burgers" on its logo doesn't make it true.
  - threethirtytwo 11mo ago
    Didn’t I say I have tons of evidence? Here’s another: https://youtube.com/shorts/zKM-msksXq0?si=bVethH1vAneCq28v https://youtube.com/shorts/zKM-msksXq0?si=bVethH1vAneCq28v Geoffrey Hinton father of AI who quit his job at Google to warn people about AI. What’s his motivation? Altruism. Man it’s not even about people saying things. If you knew how transformers and LLMs work you would know even for the most basic model we do not understand how they work.
    - mjr00 11mo ago
      I mean at a minimum I understand how they work, even if you don't. So the claim that "nobody and I mean nobody understands how LLMs work" is verifiably false.

fingerlocks 11mo ago

Try to use an LLM to a solve a novel problem or within a domain that can’t easily be googled. It will just spew over-confident sycophantic vomit. There is no attempt to reason. It’s all worthless nonsense. It’s a fancy regurgitation machine and that will go completely off the rails when it steps outside of it’s training area. That’s it.

threethirtytwo 11mo ago
I’ve seen it solve a complex domain specific problem and build a basis of code in 10 minutes what took a year for a human to do. And it did it better. I’ve also seen it fuck up in the same way you describe. So do I weigh and balance these two pieces of contrasting evidence to form a logical conclusion? Or do I pick and choose one of pieces of evidence that is convenient to my world view? What should I do? Actually why don’t you tell me what you ended up doing?