OpenAI releases 722 AI mathematics manuscripts in one go, and the mathematics community begins to review them.

Beating AI News Flash: OpenAI has released a batch of mathematical results produced by an unreleased internal frontier model, totaling 722 manuscripts, from an evaluation of approximately 4,000 open research problems. Many of the manuscripts revolve around the same core result, and when merged, they correspond to 372 groups of interrelated results.

Several of them directly touch on important long-unresolved problems. The model provided a proof of the Unique Games conjecture; proved that the Riemann zeta function has no zeros in the region where the real part is greater than 7/8, known as the "quasi-Riemann hypothesis"; and also addressed the rational Hodge conjecture on CM abelian varieties, as well as the free group factor problem that had remained unresolved since the 1940s. Among these, results such as Unique Games, quasi-Riemann, and free group factors have already been accompanied by Lean formal proofs, allowing the proof process to be checked by computer.

However, this batch of results cannot all be regarded as conclusive. OpenAI stated that the degree of verification varies across different results, and parts without formal proofs may still contain errors. The independent mathematics advisory group AGMAI, which participated in release consultations, also explicitly stated that it does not endorse these results.