The Decoder reports that after OpenAI published hundreds of mathematical manuscripts generated by AI, the Association for Human Mathematics accused the company of violating core norms of scientific research and urged mathematicians to stop working with it. Fields Medalist Terence Tao chairs the group. He shared the statement as a guest post on his blog and, on Mastodon, described a phase he calls Math 2.0, in which solving problems would no longer be the discipline’s main aim. Complexity theorist Scott Aaronson called the situation a Mathocalypse on his own blog. The association says OpenAI has already ignored the central advice of the consultative group AGMAI: advanced mathematical problems should not be tested on internal models. The statement says mathematicians did not ask for this work, and it describes the release of more than 700 files at once as a demonstration of power, not of scholarship. The Decoder also says the statement’s opening mentions the many copyright lawsuits OpenAI faces, without directly saying that the results came from training on mathematicians’ work. This story does not treat that suggestion as an established fact.
[1]
The dispute has been going on for months. OpenAI had said an internal model solved more than 100 open problems in a single month, including an approach to the Navier-Stokes Millennium Problem. After criticism grew, the company set up AGMAI at the Institute for Advanced Study. The Decoder writes that the group has no say over the pace of the company’s internal research. Aaronson contrasts OpenAI’s approach with what he calls the Anthropic model. OpenAI releases raw proof drafts in one batch, and the community then has to work through them. Anthropic partnered with two algorithm researchers whose model supplied the key idea for disproving two conjectures that were about twenty years old. The researchers were paid and wrote a version of the proof that a person could follow. Both have a cost. The first leaves the unpaid work of making proofs readable to the community. The second lets a private company choose who explains a result.
[1]The statement does not present these manuscripts as proofs that the mathematical community has independently checked and that people can already follow. The dispute is whether a problem can count as solved before a proof has been worked through on its own and made understandable. Tao and 24 other Fields Medalists had signed a statement warning of a severe misalignment between the AI industry’s goals and those of mathematics. They said problem-solving is only a tool for conceptual understanding, and that rushed announcements without proper write-ups or citations raise attribution and plagiarism questions. On Mastodon, Tao goes further: once a problem is treated as solved, it cannot be made unsolved again. This story reports the association’s and the mathematicians’ judgments. It does not verify any proof.
[1]