The Verge reported on 6 October that OpenAI released 722 manuscripts with solutions to long-standing maths problems. An unreleased model produced them. They form 372 result families, meaning groups of related papers. AGMAI, a new independent panel of mathematicians, says the batch solves hundreds of open questions. In September, OpenAI said the model had resolved more than 100 such problems. The Verge does not say how the two figures relate.
OpenAI put the papers in a GitHub repository, with rules for revisions and citations. It said it is still exploring community-hosted options that meet the committee's guidelines. The release also holds summaries of the model's reasoning, compute estimates and counts of problems attempted. OpenAI puts the typical cost of one result at three hours of ChatGPT Pro thinking. The Verge says mathematicians are still assessing the work. Debate over ethics and credit for human mathematicians continues.
In late September, AGMAI urged labs to name the model, prompts and compute costs, and to publish through academic channels where possible. It also warned against using results as marketing. A buyer can ask any vendor citing AI results the same questions. Prompts and compute costs show effort and expense, not whether a proof is right. Only experts can judge that.