I agree with Scott

Saw this interesting discussion with Scott Aaronson on TOE and I agree with many things he said. In a very short time, these AI models will be avaolavle to everyone.

One thing they discussed, where I do not agree, I think, is around the questuon of understanding. That these models have an understanding. Perhaps it can be called understanding. But more interestingly to me, I think the ability of LLMs to solve these questions says something more about language. All the problwms and proofs and stated in a languge. And this language is somehow finite. You can express these statements in a finite amount of words. And they have some kind of, abstract yes but, logic to them. The follow logic structure, and therefor they can be found and expressed in a language by LLMs as they are good at predicting streams of words based on some structure.

It seem to me there is an insight here. Perhaos not. Who konws...