On October 6, OpenAI dumped 722 preprints purportedly written by an internal AI model onto GitHub, claiming progress on 372 open problems across algebra, geometry, and logic — including results touching three Millennium Prize Problems. Less than 24 hours later, a single sign error cascaded into three manuscript withdrawals and 14 revisions. The release has ignited a storm in global mathematics. Terence Tao, regarded as one of the world's living math legends, warned of "proof indigestion" from AI-generated manuscripts. Over 8,000 researchers have endorsed an open letter accusing OpenAI of rushing the unvalidated work — a pattern critics say echoes the controversial September 8 Navier-Stokes announcement. The Association for Human Mathematics issued a direct boycott call, framing the simultaneous release as "a demonstration of power, not scholarship."
Key Context: This is not OpenAI's first foray into automated math research, but it is by far the most ambitious and the most contentious. The company's GitHub release affected problems including the Hodge conjecture, the Riemann hypothesis, and the Birch and Swinnerton-Dyer conjecture — each carrying a $1 million Clay Millennium Prize. Roughly half the results were released without Lean formalization verification, and OpenAI's own advisory group (AGMAI) has privately recommended stricter release protocols. The incident has become a pivotal case study in how AI-generated research challenges traditional norms of peer review, authorship credit, and scientific attribution.
Top comments (0)