Current outcome
OpenAI published 722 mathematics manuscripts generated by an unreleased frontier model, spanning 372 result families and claiming progress on several long-standing open problems; some results have Lean formal proofs, but verification is inconsistent, AGMAI has not endorsed them, and the mathematics community is reviewing them.
Progress timeline
3 material updates- #01
OpenAI drops another batch of mathematical breakthroughs
OpenAI released a batch of 722 manuscripts spanning 372 result families containing solutions to long-standing mathematics problems generated by an unreleased frontier model, continuing a run of AI-driven math breakthroughs that has stirred debate over research ethics.
Source evidence: The Verge AI
- #02
OpenAI一夜甩出722篇数学论文！黎曼霍奇BSD全上阵，数学家：读不过来
OpenAI released 722 mathematics papers covering major open problems including the Riemann Hypothesis, Hodge conjecture, and BSD conjecture, drawing pushback from three Fields medalists who stated the release does not constitute their endorsement.
Source evidence: QbitAI 量子位
- #03
OpenAI一口气放出722篇AI数学手稿，数学界开始验收
Additional details emerged: the manuscripts address specific results including the Unique Games conjecture, a quasi-Riemann hypothesis, the rational Hodge conjecture for CM abelian varieties, and the free group factor problem; some have Lean formal proofs, but verification is inconsistent and the independent math advisory group AGMAI has not endorsed the results.
State after update: OpenAI published 722 mathematics manuscripts generated by an unreleased frontier model, spanning 372 result families and claiming progress on several long-standing open problems; some results have Lean formal proofs, but verification is inconsistent, AGMAI has not endorsed them, and the mathematics community is reviewing them.
Source evidence: theblockbeats