Courant Institute of Mathematical Sciences

Actually, Anthropic wasn't working on it.

A mathematician who works for Anthropic alongside a mathematician at Courant at Courant Institute was working on it.

And they were actually using a mixture of models, both Entropic and OpenAI.

So anyway, OpenAI then dedicated a tremendous amount of compute to ultimately solve a version of this problem, right? So a much more stronger version than what the entropic NYU, let me just continue.

So there's two issues I wanted to drill down, which might have resonance with the broader AI community.

The first one is directly involves this controversy, which is basically if you're working on a problem, should you have some assurance from the model provider that your work is basically preserved, I mean, private to your efforts, right? So the reason there's some suspicion is if you know something about this problem, there's a lot of different directions that people were trying to attack this problem.

And it just happened to be the case that the people at NYU Courant Institute, which is, by the way, the leading center probably for this problem in math, and the person at Anthropic chose a path that was, let's just say, not as common But then at the end of the day, after the OpenAI announced their results and people waded through it, it seemed like they chose the same direction.

So there's some suspicion.

They briefed the result on a press call without publishing the full mathematical argument.

Tristan Buckmaster, the NYU Courant mathematician closest to this problem, confirmed he hasn't reviewed it either.

A claim is not a result until somebody can check it.

And the comparison is unflattering, because Buckmaster and his collaborator, Levent Alpog, released lean formalized proofs of related results.