OpenAI publishes 722 math manuscripts from frontier model — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI released 722 manuscripts covering 372 result families that group related papers, including solutions to "hundreds" of open mathematics questions produced by an unreleased frontier model.
- OpenAI said in September its model had "resolved more than 100 long-standing open problems across most areas of mathematics," with the released papers showing the average result used the equivalent of three hours of ChatGPT Pro thinking.
- AGMAI (the Advisory Group on Mathematics and Artificial Intelligence) published first guidance in late September urging AI labs to release results through established academic channels and disclose model names, prompts, and compute costs.
- AGMAI implored labs to "refrain from treating the release of mathematical results as marketing vehicles to promote their models," calling such practices a source of significant harm to the mathematical community.
- OpenAI published the manuscripts in a GitHub repository with protocols for paper revisions and citations, and committed to improving citation, mathematical exposition, and presentation in future releases.
- OpenAI's batch adds to a growing body of AI-generated mathematical results from rival labs like Anthropic, including results concerning a Millennium Prize problem among the field's most famous open questions.
- Mathematicians are fiercely debating research practices, ethics, and how AI companies credit the human mathematicians whose work the systems build upon and potentially use to generate results.
Why it matters: AGMAI's public framing of these releases as potential "marketing vehicles" signals organized community pushback on OpenAI's handling of the announcement, with the unresolved question of how labs credit the human mathematicians whose work the models build upon now sitting at the center of a discipline-wide ethics debate.
Ask SkimNews
