OpenAI drops another batch of mathematical breakthroughs

OpenAI has released a batch of 722 manuscripts containing solutions to hundreds of long-standing open mathematics problems, generated by an unreleased frontier model. Covering 372 result families, this release extends a series of breakthroughs that have impressed and unsettled the mathematical community. The newly formed independent advisory group AGMAI confirmed the inclusion of solutions to "hundreds" of open questions. OpenAI had previously announced in September that its model resolved over 100 open problems. The papers include reasoning summaries, compute estimates, and problem attempt statistics, with the average result reportedly using the equivalent of three hours of ChatGPT Pro thinking. AGMAI has urged AI labs to release results promptly through academic channels, disclose details, and avoid using releases as marketing. OpenAI published the results on GitHub with protocols for revisions and citations, and committed to improving future releases. The full impact will take time to assess, adding to a growing body of AI-generated mathematical results that has provoked debates over research ethics, credit, and practices.

(Source：The Verge)