‘Slop mathematics’: OpenAI walks away from a Caltech AI maths contest

OpenAI has pulled its $1m sponsorship of a student-run AI maths event at Caltech. Mathematicians had signed an open letter calling for the event to be scrapped. The next day, 25 Fields Medallists published a separate declaration of their own.

Two letters in two days, from two different groups, is the sharpest response the mathematical community has made to AI companies so far. Neither is about whether the models work. Both are about what the companies do with the results.

The first letter, and the word that stuck

A coalition of current and former Caltech mathematicians circulated the open letter from 8 September and published it on the 10th. It carried 771 signatories at publication. Business Insider put the figure above 700 academics and students. Only the Caltech community and mathematics researchers at PhD level or above may sign it.

The letter asks the organisers to suspend the event. It gives five grounds. The first is the one the coverage picked up. The letter calls it slop mathematics: produced fast, and left for other people to check later.

Its central charge is blunter than the label suggests. Research mathematicians end up verifying, spreading and correcting the results AI companies announce. That labour “goes uncompensated, uncredited, and unacknowledged”, the letter says. It concludes that AI companies are engaging in research misconduct.

One line sets out the economics. No reasonable future model of mathematical research gives every research mathematician $20,000 in AI credits to prove a result, the letter says.

OpenAI walked, Anthropic did not

OpenAI withdrew on Thursday 10 September. Dan Roberts, an OpenAI research lead, wrote on X that the company recognises the rapid progress of AI in mathematics is disruptive. He said OpenAI wants to engage the maths community on how to integrate the technology. Truman Dickerson reported the withdrawal for Business Insider.

The company took back its $1m in credits. Anthropic put in the other $1m and has not withdrawn. Business Insider said Anthropic did not immediately answer its request for comment.

Anthropic spent the same fortnight courting mathematicians rather than leaving them. TNW reported on 6 September that it used Claude to formalise Fermat’s Last Theorem in Lean in 11 days. It has since widened free access and research credits for formalisation work.

The second letter came from the top of the field

On 11 September, 25 Fields Medallists published a declaration titled A Severe Misalignment of AI in Mathematics. The Economist reported the number as 24. The site itself lists 25, with medals awarded between 1978 and 2026, and Terence Tao described himself as one of 25 initial signatories.

The signatories include Tao, Peter Scholze, Maryna Viazovska, Cédric Villani, Pierre Deligne, Manjul Bhargava and Martin Hairer.

Their argument is narrower than the Caltech letter and aimed higher. Solving problems is only a tool and a proxy, they write. The primary goal is conceptual understanding and insight. They warn that mass production of true or false statements at increasing pace “could destroy fertile ground”.

The declaration never names the Mathathon and asks for no cancellation. It is a statement about benchmarks.

What the event actually is

The organisers announced the Mathathon on 4 September. It runs at Caltech from 30 October to 1 November. One hundred teams get 40 hours each, and roughly $20,000 in credits, to produce a result on an open problem.

Teams whose work the judges rate as worth the community’s attention go into a second round lasting six months. That round now requires an arXiv preprint, a video presentation and an auto-formalised solution. There is also an educational track for course notes, explainers and cleaner proofs of existing theorems.

The organisers had over a thousand applications by 10 September, from undergraduates through to faculty. Alongside the AI credits they list support from DARPA expMath and Cognition.

The organisers are undergraduates, and they said no

The five named organisers are students. They published a response the same day as the letter, conceding that several of the concerns are valid and that their original description did not address them.

They made changes. They added the arXiv requirement after the letter asked for it, and they expanded the educational track. They also cut a line from the website asking what the role of a mathematician is when AI can solve conjectures faster. The letter’s own footnote records that removal.

They did not cancel. All the organisers have signed the Leiden Declaration, the formal statement mathematicians issued in June telling AI companies to stop using their work without permission.

Caiman Moreno-Earle, a 19-year-old Caltech sophomore, started the event. He told Business Insider he asked Anthropic for $1m on a call, expecting to be offered $50,000. He said younger mathematicians cannot wait for the rules to settle. “We don’t have the luxury of waiting around to find out,” he said. Organisers are now seeking credits to replace OpenAI’s.

What set this off

The trigger was the week before. OpenAI said on 8 September that a swarm of its agents had produced a solution to the Navier-Stokes problem, and published the proof the next day while saying it would not claim the $1m Millennium Prize.

Tristan Buckmaster of NYU, who was working the same route with Levent Alpöge of Anthropic, said OpenAI had offered him authorship arrangements that excluded Alpöge. Sébastien Bubeck of OpenAI denied trying to cut Alpöge from credit. Buckmaster has since told the ABC that racing to publish is now pointless, and that the game is up.

TNW reported in August that the AI maths results most often cited are counterexamples rather than proofs. That distinction is what both letters are defending.

Unresolved

Three things are open. Whether the Mathathon finds replacement sponsors, and on what terms. Whether OpenAI’s withdrawal changes how it approaches the field, or only this event. And whether the Clay Mathematics Institute, which still lists Navier-Stokes as unsolved, accepts the proof at all.