Grant Harvey

But kind of like going through after the fact and seeing what it did.

Yeah.

Obviously, IMO happened earlier this year.

That was the International Mac Olympiad.

And both you and Google were competing in that.

And they did eventually release, I believe, a version of that model that they ran IMO with called DeepThink.

And I think there's a version of that that certain users of Google can use now.

And I don't know if you mentioned that you're using the same IMO model or version of that that you used for this competition.

Do you think OpenAI will eventually release a similar reasoning model from the ICPC system available in production in ChatGPT, perhaps as a future model?

Or what are your thoughts there?

I am sure like being able to like take a historic victory is a good fruit of your labor.

Um, I guess like switching gears just a tiny bit here.

There's been buzz about the new version of Codex and, you know, a lot of people have been talking about how it can run for seven hours straight in some instances.

How do you think about like scaling, you know, from a five hour ICPC problem to these more longer horizon agent tasks, like in terms of what you're going to be doing next, you know?

No, we both clearly sparked something.

Yeah, on the outside, you know, I'm just like, that's cool, but I don't understand it.

I actually, my question is kind of related to what Corey's saying.

So like when it comes to tool use, I've always wondered like how exactly, and you know, you can just kind of like, you can talk about this at whatever level of abstraction surface level you want, but how is it that the AI is actually able to use tools?

Is it just like getting the responses of the feedback, like streamed back to it?

Like that part still kind of blows my mind how that works.