Post

The more lurid accusations aside, this seems particularly relevant to me. Really calls into question the “pure AI” capabilities claimed in past announcements. Possibly industry-wide. Of course, no way of directly auditing it.

To be sure, I don’t really buying the Buckmaster insinuation that a hastily-assembled research strike force could suddenly complete a general solution, if indeed OpenAI has it. Probably, that background work has been ongoing for some time at OpenAI and Anthropic. In the past (?), at DeepMind.

but also, the OpenAI crew were not mathematicians, it remains possible that OpenAI’s models actually discovered a real breakthrough where Claude had to rely on human ingenuity. Hard to say…

OpenAI enters the chat they heard that Anthropic may have made progress, so Sebastien Brubak put together a research team and pursued it with an internal model said NYU prof is accusing OpenAI of trying to take credit, which kind of seems to be the case officechai.com/ai/mathemati...

Levant (Ant employee) and Tristan Buckmaster (NYU prof) worked together using not just Claude but also Sol & (more recently) Astra. however, Claude did produce very real and probably valuable work. Granted, Tristan describes it as barely readable “AI slop” oh Claude…

from what i can tell, the millennium prize problem has still not been claimed to have been solved. Rather, a tremendous amount of progress has been made in an extremely short period of time, causing Tristan to label it, “a Deep-Blue—Kasparov moment” for mathematics