OpenAI Drops 722 AI-Generated Math Manuscripts

OpenAI released 722 AI-generated math manuscripts on GitHub, claiming progress on unsolved problems despite mathematicians' criticism.

@sharedot

OpenAI Drops 722 AI-Generated Math Manuscripts

By @sharedot · · 6 pages

  • AI
  • Mathematics
  • Openai

OpenAI released 722 AI-generated math manuscripts on GitHub, claiming progress on unsolved problems despite mathematicians' criticism.

A batch release without peer review

OpenAI on Tuesday published 722 mathematical manuscripts produced by an unreleased frontier model, organized into 372 result families and posted to GitHub under an Apache-2.0 license, BigGo Finance reports.

The model was posed roughly 4,000 problems during the evaluation, and the average result consumed compute equivalent to about three hours of ChatGPT Pro thinking time. Many proofs were formalized in Lean for machine verification, though OpenAI cautioned that unformalized results may contain errors.

The release follows the company's September announcement that an internal model had resolved more than 100 open problems, and comes directly to GitHub rather than through traditional peer review.

Claims that stunned number theory

KuCoin's report goes further, claiming the model proved the "quasi-Riemann hypothesis": that Dirichlet L-functions have no zeros in the half-plane Re(s) > 7/8, eliminating the Landau-Siegel zero — a breakthrough it says was verified in Lean, though it stops short of the full Re(s) = 1/2 line.

BigGo Finance describes a zero-free region for the Riemann zeta function at Re(s) > 11/12 and a proof of the Hodge Conjecture for CM abelian varieties, noting the zeta write-up was edited by a human for readability.

Ten reasoning summaries accompany the release, spanning analytic number theory to operator algebras.

Mathematicians push back on the process

Much of the criticism targets the manner of release rather than the mathematics.

Northwestern University mathematician Bryna Kra said "math by tweet and math by press release" is not how to nurture the ecosystem the labs trained on, and that an August request to publish absorbable papers instead of a single batch was ignored, according to BigGo Finance.

The Advisory Group on Mathematics and AI, consulted on the release, urged labs to avoid using math results as marketing and does not endorse testing advanced problems on proprietary models the community cannot access.

Enormous compute and what comes next

The scale of the underlying effort is striking: BigGo Finance reports OpenAI's coordinating agents sent 4.9 million messages and consumed roughly 300 billion output tokens, with the group that resolved the fluid problem reaching about 10,000 concurrent agents, hitting its result about 88 hours after launch and needing 17 more hours for Lean verification.

OpenAI pledged to fund workshops and programs to help mathematicians digest the results, while AGMAI said publication is the beginning, not the completion, of turning machine output into human mathematical knowledge.

Sources

  1. finance.biggo.com › OpenAI Releases 722 AI-Generated Math Manuscripts, Claims Breakthroughs on Hundreds of Unsolved Problems
  2. kucoin.com › OpenAI Solves 722 Math Problems, Quasi-Riemann Hypothesis Proven

More on AI