Comment by alfor
3 years ago
I don’t know what insight we expect of Chomsky at this point.
He don’t seem to understand how it’s going and where it’s going.
I at this point AI is only limited by our capacity to create memeable flaws.
If you can create a criticism of ChatGPT that is concise, accurate and funny it will go viral and get fixed fast.
Yes at the moment it’s intelligent is very wide but not that deep (brute force allusions) that will get fixed and it will be way more efficient at the same time (more compression of information) It doesn’t have real experience/connection to our world: expect that to change with video/audio information and robotic manipulator. It say falsehood, doesn’t know what it doesn’t know: actually it’s in the API but not exposed in chatGPT. Expect that to get fixed also. Morality is based on iteratively playable games, that can get baked into it also.
This is what I've been repeating for months/ years. Chomsky had some interesting theories, that for a while, were very worth discussing as frameworks for the emergence of language.
Now we have chatGPT, a very very interesting framework for the discussion of emergence and language. And even more dramatically, it is in some sense empirical. We haven't yet even begun to explore it, but this imo is the allegorical to the discovery of DNA in the context of the theory of evolution.
Before Watson and Crick and Franklin, we had a coherent theory of evolution (ish). We knew all about selective breeding and it was pretty clear that descent and the transmittance of information 'happened'. Mendelian genetics was enough for that. But as useful as a teaching tool like Mendelian genetics is, the entire world changed with the discovery of the actual-particle responsible for that information. The world changed with the discovery of DNA. I don't know the zeitgeist of other competing theories for how that information was transmitted. But what we do know now, is that they were all wrong, to the extent that they don't get mentioned or discussed.
A real interesting discovery extending from ChatGPT is the apparent emergency of language from what amounts to large piles of information and sufficient complexity. It appears that Chomsky may just be entirely wrong.
Fantastic! AGI solved in one go. While you’re at it, could we also have cure for cancer too by the end of the year please?
I raise you the challenge:
create a memeable flaw of chatGPT that won’t be fixed by the beginning of 2024.
I’ll do my best on the cancer front ;)
And here I thought raising the challenge meant increasing the difficulty. I must be hallucinating!
Jokes aside; of course the memeable (internet famous) flaws will be fixed by RLHF. But that means nothing in terms of intelligence, does it?
> He don’t seem to understand how it’s going and where it’s going.
Well... since you and Noam are at odds, logically it's either: (1) you; (2) him; (3) or both who don't understand.
And, frankly, since you claim this:
> at this point AI is only limited by our capacity to create memeable flaws
I think we're already down to (1) or (3).
He's 94, lets just be impressed he can have an opinion at all