Leadership1 publisher3 min readPublished
OpenAI pulls out of a Caltech AI mathathon after 700 academics call for its cancellation
Two labs put $1 million each of computing power behind a student event that asked what mathematicians are for. One has since withdrawn, and the 19-year-old who pitched it is still dealing with the objections.
The Board Room · Leadership desk
What happened
- A Caltech sophomore, Caiman Moreno-Earle, organised a "mathathon" in which participants would use AI to attack some of the world's hardest theoretical problems in a single multi-day session.
- More than 700 academics and students then signed an open letter denouncing the event as promoting AI "slop" and calling for it to be canceled.
- OpenAI withdrew its support for the mathathon in the face of that backlash, while the event itself remains scheduled with some alterations.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- exposure A sponsor can exit inside a news cycle. The undergraduates keep working with the professors they asked to judge the event, some of whom they had consulted privately before those consultations turned into a public letter.
- constraint Only frontier labs give away compute at the scale AI agents on open problems require, so the people best placed to test the labs' claims about mathematics are dependent on the labs to run the test at all.
- contradiction The signature list does not support a young-versus-tenured story: Business Insider describes it as academics and students. The letter's central objection is empirical: that the faster-than-humans claim is "empirically dubious."
- precedent Lab sponsorship of an academic event now comes with a demonstrated exit at the first reputational cost, and the next set of organisers has to plan for a funder leaving mid-project.
Moreno-Earle told Business Insider the organisers went into a Zoom meeting with Anthropic and asked for $1 million: "We were like, OK, look, we're gonna pitch some absurdly high number," he recalled. "We'll be happy when they give us like 50k." [12] The sum they would have been satisfied with was one twentieth of the sum they named [1]. Anthropic agreed to the full $1 million in computing power, and OpenAI added another million soon after [13].
When the open letter passed 700 signatures, OpenAI backed out [5], and half of the disclosed compute went with it [2]. Donated compute can be taken back. Moreno-Earle said some of the people the organisers had been consulting privately about responsible AI use went public with the letter calling for cancellation [17]. The organisers had also asked Caltech professors to judge the event and worked with them on a framework for AI use [15].
The compute that made a multi-day AI mathathon possible arrived with two sponsors' names attached [13]. Those names are among the things the letter objected to [18].
"As young mathematicians, we don't necessarily have some of the same luxuries that these older mathematicians do," Moreno-Earle said. "We look to the future, and we don't even know if there's going to be grad school open positions by the time we get to Ph.D." [7] Students signed the letter too, though: Business Insider describes it as signed by more than 700 academics and students [5].
The flyer that went live on September 4 [16] asked, "What is the role of a mathematician when AI can solve conjectures faster?" [19] Signatories answered that "The claim that AI can solve conjectures faster than human mathematicians appears both motivated and empirically dubious" [20], and that the event as sponsored by Anthropic and OpenAI "is likely to have destructive impacts for the mathematical community" [18]. They also raised fears that the labs wanted a garish marketing opportunity and that the format would incentivise rushed, poorly understood AI-generated work [22].
AI's work in math is often quick and dirty and takes hours of human-led research to make sense of, the professors told Business Insider [8]. "The actual progress is when human knowledge advances," said Alvaro Lozano-Robledo, a professor at the University of Connecticut [9].
Business Insider reports that fault lines between mathematicians and AI companies opened last week after OpenAI said it had solved a problem so difficult that a math institute was offering a standing $1 million prize for the solution [21]. A lab making that claim in public, then funding an event asking what mathematicians are for, gave the signatories the claim they attack. Whether AI displaces working mathematicians is a question for the decade; the question in front of Moreno-Earle this term is whether a 19-year-old sophomore [1] can run a judged event in his own field after a sponsor withdraws. The event is still set to go ahead with some slight alterations [6].
What to watch
- Whether Anthropic keeps its $1 million of computing power in place through the event now that OpenAI has gone.
- What the "slight alterations" to the format are, and whether the Caltech professors asked to judge stay on.
- Whether any mathathon result survives the hours of human-led verification professors say AI output requires.