Vibeleaderboard
← All Intel
Intel / video

OpenAI’s IMO Team on Why Models Are Finally Solving Elite-Level Math

Source
youtube.com
Author
Sequoia Capital
Date

Context

Olympiad proofs from an AI are hard to check, because there is no answer key. In a Sequoia Capital conversation, three OpenAI researchers behind a gold-medal-level International Mathematical Olympiad result, including Noam Brown, said they hired external former IMO medalists to grade the solutions. Each proof was graded by three medalists, who reached unanimous agreement on correctness. This is the team's own account of its .

They said the proofs are hard to read because the team had not optimized for readability, and they posted the originals for transparency. On problem six, which the interviewer described as usually the hardest, the model returned no answer despite heavy compute instead of inventing one. Brown contrasted that with earlier reasoning models, which mathematicians told him could give convincing but wrong answers. The claim supported is narrower than the headline: expert-graded proofs on attempted problems, plus a willingness to decline one.

Terms in this piece · Glossary
  • Model — A computational system that uses learned patterns to produce an output from information supplied to it.
  • Process — A running instance of a program, with its own working memory, open files and identity.
  • test-time compute — Spending more computation when the model answers — thinking longer, trying multiple attempts — to buy accuracy without training a bigger model.
Read the source www.youtube.com
More from Sequoia Capital
Recommended reads
Comments

Checking sign-in…

Loading comments…