Terence Tao

Type of scaling to once you solve one problem to make the AI attack 100 adjacent problems.

The things that humans do still, so where the AI really struggles right now is knowing when it's made a wrong turn.

that it can say, oh, I'm going to solve this problem.

I'm going to split up this one into these two cases.

I'm going to try this technique.

And sometimes if you're lucky, it's a simple problem.

It's the right technique and you solve the problem.

And sometimes it will get, it will have a problem.

It would propose an approach which is just complete nonsense.

but it looks like a proof.

This is one annoying thing about LLM-generated mathematics.

We've had human-generated mathematics that's very low quality, like submissions of people who don't have the formal training and so forth.

But if a human proof is bad, you can tell it's bad pretty quickly.

It makes really basic mistakes.

But the AI-generated proofs, they can look

superficially flawless.

And it's partly because that's what the reinforcement learning has trained them to do, to produce text that looks like what is correct, which for many applications is good enough.

So the errors are often really subtle, and then when you spot them, they're really stupid.

Like no human would have actually made that mistake.

Yeah, so the sense of smell.