OpenAI Claims Major Progress on a Second Millennium Prize Problem, Sparking Academia Backlash

新智元 · wechat · 2026-09-17

OpenAI president Greg Brockman says the company has made "significant progress on another" Millennium Prize Problem, following its Navier–Stokes breakthrough, fueling speculation of a DevDay reveal — with rumors pointing to two of the six remaining problems, possibly including P vs NP.

The earlier breakthrough used 10,000 reasoning agents over 88 hours, 2.7M messages, and 130B output tokens to prove smooth fluids develop finite-time singularities, fully formalized in Lean.

Controversy: After hundreds of mathematicians protested, OpenAI withdrew sponsorship of Caltech's first Mathathon. NYU's Tristan Buckmaster alleged his pre-publication progress leaked to OpenAI, which then burned 300B tokens in about a week to produce a complete proof with an unreleased model — and claims mathematician Sébastien Bubeck pressured him to drop a co-author's name. Brockman says the solving model's training cutoff predates the scholars' publication.

The core anxiety: when a scholar's public AI opens a crack in a century-old problem, a giant with 1000× compute can flatten the rest in days. Brockman argues these problems are the ultimate testbed for long-chain rigorous reasoning that will transfer to drug discovery and molecular dynamics.

Related event: Report: OpenAI nears solving the Hodge Conjecture(13 posts)→

Original post →

More from Companies & People

Companies & People channel →