Terence Tao

I mean, there's still a problem with hallucinating references that don't exist.

But this, I think, is a civil world problem if you train in the right way and so forth.

and verify using the internet.

You should, in a few years, get to the point where you have a lemma that you need and say, has anyone proven this lemma before?

And it will do basically a fancy web search AI-assisted and say, yeah, there are these six papers where something similar has happened.

And

I mean, you can ask it right now and it will give you six papers of which maybe one is legitimate and relevant.

One exists but is not relevant and four are hallucinated.

It has a non-zero success rate right now, but there's so much garbage, the signal-to-noise ratio is so poor that it's...

It's most helpful when you already somewhat know the literature, and you just need to be prompted to be reminded of a paper that was already subconsciously in your memory.

Yeah, that's...

Yeah, that it can sometimes do, but when it doesn't, it's buried in a list of options to which the other- They're bad.

Yeah.

Yeah, yeah, yeah.

There's a big hump to overcome right now.

I mean, it's like self-driving cars.

The safety margin has to be really high for it to be feasible.

So yeah, so there's a last mile problem with a lot of AI applications that they can develop tools that work 20%, 80% of the time, but it's still not good enough.

And in fact, even worse than good in some ways.

Right, right.