OpenAI Says It Solved a Famous Math Problem. The Harder Problem Is Who Gets Credit.

Two days after OpenAI announced that its models had proved a famous unsolved math problem, roughly 150 students, postdocs, and professors filled a UC Berkeley classroom to hear Ken Ono tell them what comes next. The talk was scheduled for 50 minutes. With questions, it ran past two hours.

Ono, a number theorist who left academia for the AI start-up Axiom Math, did not soften the news. "You might be graduating into a profession that might not even exist, or that will be very different than what you expected," he told the room. A student near the front, hands trembling, asked Ono: "What is your very best?" The exchange captured the meeting: a generation being told its vocation has been reset, asking the person delivering the message for something to hold on to.

The trigger sits in plain public view. On September 8, OpenAI said its models had proved a famous decades-old problem, reigniting an argument that had been building all summer as AI systems regularly produced proofs of long-standing conjectures. Within days, Scott Aaronson, one of the field's most cited theoretical computer scientists, posted a response that read like a eulogy for a profession. "In whatever years I have left, I don't expect that I'll ever again prove a theorem because I'm actually needed to prove it," he wrote on his blog. Then, a line that has since circulated well beyond mathematics: "Human mathematicians are forevermore dethroned as the main theorem-proving entities on planet earth."

The Berkeley audience was not buying the obituary, but it was not buying the press release either. Several students challenged Ono directly on what they called the "shameful way that AI companies are treating mathematics." The room's mood, as one reporter in the room observed, ran through anger, frustration, grief, confusion, and fear, sometimes in the same answer. Friends and family outside the field, the reporter noted, did not see the problem at all, a social gap that may matter as much as the technical one.

The conflict is not really about whether a machine can produce a proof. Most mathematicians already accept that it can. The fight is about what a proof is for, who gets credit when a machine finds one, and what human mathematicians will be paid and trained to do once the routine part of theorem-hunting is automated. Those are authorship questions, and they are not settled. They are also questions the math profession has been unusually slow to face in public, even as the tools land in its seminar rooms.

If a model can clear a proof that used to take a career, more mathematicians can spend their years on the harder second step: deciding which problems are worth solving in the first place, interpreting what a proof means, and teaching the next cohort how to think under new rules. The work does not disappear. It moves up the stack, into the choices that no current model is being trained to make.

The pro-progress case is also real, but it cuts both ways. The same companies selling the speed-up are not, today, helping set the credit rules, the disclosure norms, or the authorship standards that the field will live under. That is the part Ono was pressed on, repeatedly. Several students asked him what Axiom Math, and the labs like it, owe the community whose training data and open questions built the field in the first place. Ono did not have a clean answer, and he said so. The room stayed anyway.

Mathematics is not ending. The Quanta reporter's observation that the wider public does not yet see the disruption is itself a finding: the profession is renegotiating its own future in front of a small audience, and the rules it lands on will shape what "mathematician" means for the next cohort. The trembling-hands question in the Berkeley room is the right one, and it is still open. "What is your very best?" is a fair thing to ask of the people building the systems, and of the people deciding what those systems get to author.