6 ms·

> We have no good definition for what sentience is either; maybe our brain is word association shenanigans; This is just stupid. We've pumped more English thr

> We have no good definition for what sentience is either; maybe our brain is word association shenanigans; This is just stupid. We've pumped more English through GPT-3 than any existing English-speaking human has absorbed in their entire lifetime, and what we've ended up with is something that very very clearly has so utterly failed to generalize even the most basic level of understanding of basic concepts that if you ask it to count the number of letters in a word it will cheerfully pump out the wrong answer ("there are thirteen letters in the word 'twelve'") because its dataset correlates the two words with one another and it has learned precisely sweet fuck all about what it means to count, something a child's brain picks up with exposure to many orders of magnitude fewer language examples. To imagine your brain is just an LLM is to mistake your reflection for another person in the room. Utterly daft. Get off the LLM hype train and start looking at these things objectively and critically. They're nowhere near what you're imagining.

fiso64 4y ago
>if you ask it to count the number of letters in a word it will cheerfully pump out the wrong answer ("there are thirteen letters in the word 'twelve'") Not to be too blunt, but you seem to be talking out of your ass. I'm tired of the overly dismissive comments here on hackernews by people who didn't bother to do the bare minimum of research. LLMs do not work with individual characters, they use tokens (i.e multiple characters, or sometimes entire words).
- anonzzzies 4y ago
  Yep. It’s not called tokens for nothing.
- BaseballPhysics 4y ago
  Wow, you... completely misunderstood the purpose of that example. Seriously, slow down and read it again, all the way to the end of that sentence.
  - fiso64 4y ago
    Uh, no.. I don't believe I did. Asking it to count the letters in a word is like asking a human to listen for a 30kHz tone without them knowing that humans can't hear it. I'd expect a lot of false positives.
    - bonsaibilly 4y ago
      Which reinforces the point I was making. Stop trying to win an argument online and think for 5 seconds about what it means for the claim "the human brain is just an LLM" if you're arguing that an LLM is naturally ill-suited to this task human children can do without issue.
      - fiso64 4y ago
        >the human brain is just an LLM Never have I said such a thing. Also, BPEs aren't a natural limitation of any transformer-based model, it's a trick to save compute for LLMs.
        BaseballPhysics 4y ago
        So what is this supposed to mean? > We have no good definition for what sentience is either; maybe our brain is word association shenanigans; Sure sounds like you're suggesting the brain is an LLM to me, and I can't blame bonsaibilly for thinking that.
        fiso64 4y ago
        Keeping track of different usernames is hard, and I can't blame you for confusing me with the other guy :) snark aside, I think he does have a point - afaik, we don't know what intelligence is, so it's kinda hard to make any argument about fundamental differences between "true" intelligence and LLM intelligence. I do feel like there should be something else. I saw somebody describe their mind as consisting of a "babbler" and a "critic" (a GAN, basically). The LLM would be the babbler while the critic is not yet implemented, and this sounds intuitively right to me. Then again, my intuition could be completely wrong and we may be able to get to human level intelligence with further scaling. I haven't seen any solid counterarguments yet. And not even the biggest LLM believers are denying the fact that it's not exactly the same thing as a human brain, but the question is whether it captures the gist of it.
williamcotton 4y ago
Here is an objective and critical assessment of large language models and how to empirically improve the results of questions involving calculations like the lengths of strings: https://github.com/williamcotton/empirical-philosophy/blob/main/articles/from-prompt-alchemy-to-prompt-engineering-an-introduction-to-analytic-agumentation.md https://github.com/williamcotton/empirical-philosophy/blob/m...

fiso64 4y ago

>if you ask it to count the number of letters in a word it will cheerfully pump out the wrong answer ("there are thirteen letters in the word 'twelve'") Not to be too blunt, but you seem to be talking out of your ass. I'm tired of the overly dismissive comments here on hackernews by people who didn't bother to do the bare minimum of research. LLMs do not work with individual characters, they use tokens (i.e multiple characters, or sometimes entire words).

anonzzzies 4y ago
Yep. It’s not called tokens for nothing.
BaseballPhysics 4y ago
Wow, you... completely misunderstood the purpose of that example. Seriously, slow down and read it again, all the way to the end of that sentence.
- fiso64 4y ago
  Uh, no.. I don't believe I did. Asking it to count the letters in a word is like asking a human to listen for a 30kHz tone without them knowing that humans can't hear it. I'd expect a lot of false positives.
  - bonsaibilly 4y ago
    Which reinforces the point I was making. Stop trying to win an argument online and think for 5 seconds about what it means for the claim "the human brain is just an LLM" if you're arguing that an LLM is naturally ill-suited to this task human children can do without issue.
    - fiso64 4y ago
      >the human brain is just an LLM Never have I said such a thing. Also, BPEs aren't a natural limitation of any transformer-based model, it's a trick to save compute for LLMs.
      - BaseballPhysics 4y ago
        So what is this supposed to mean? > We have no good definition for what sentience is either; maybe our brain is word association shenanigans; Sure sounds like you're suggesting the brain is an LLM to me, and I can't blame bonsaibilly for thinking that.
        fiso64 4y ago
        Keeping track of different usernames is hard, and I can't blame you for confusing me with the other guy :) snark aside, I think he does have a point - afaik, we don't know what intelligence is, so it's kinda hard to make any argument about fundamental differences between "true" intelligence and LLM intelligence. I do feel like there should be something else. I saw somebody describe their mind as consisting of a "babbler" and a "critic" (a GAN, basically). The LLM would be the babbler while the critic is not yet implemented, and this sounds intuitively right to me. Then again, my intuition could be completely wrong and we may be able to get to human level intelligence with further scaling. I haven't seen any solid counterarguments yet. And not even the biggest LLM believers are denying the fact that it's not exactly the same thing as a human brain, but the question is whether it captures the gist of it.

williamcotton 4y ago

Here is an objective and critical assessment of large language models and how to empirically improve the results of questions involving calculations like the lengths of strings: https://github.com/williamcotton/empirical-philosophy/blob/main/articles/from-prompt-alchemy-to-prompt-engineering-an-introduction-to-analytic-agumentation.md https://github.com/williamcotton/empirical-philosophy/blob/m...