OpenAI walks out of Caltech’s mathathon over “slop” letter

OpenAI on September 10 dropped its sponsorship of Caltech’s mathathon, a student-run event that OpenAI and Anthropic had backed with $2 million in AI credits. The exit came a day after current and former Caltech mathematicians published an open letter saying the event “is likely to have destructive impacts for the mathematical community.”

Dan Roberts, OpenAI’s research lead, announced the decision on X. “We recognize that the rapid progress of AI in mathematics is disruptive,” he wrote. “We’re looking to engage with the math community more on the best way to integrate this technology and communicate its impacts.” Business Insider reported that it was unclear how much of the $2 million had come from OpenAI, or what happens to those tokens now.

What the mathathon was meant to be

The event was announced in early September as “the first hackathon ever devoted to research level mathematics.” Organizer Brian Zhao, a junior studying applied mathematics at Caltech, said he and his roommate came up with the idea earlier in the summer: lock themselves in a basement and try to solve math problems for 40 hours straight, the way programmers and designers do at hackathons.

The format scaled that up. In the first round, each team gets 40 hours and $20,000 in tokens to work on an open problem with large language models. Teams whose answers are “promising and well explained” move to a second round, where they get six months and more AI tokens to keep going. The application page described the event as a place for young mathematicians to “responsibly use AI tools to augment human understanding of mathematics.” Many of the organizers are undergraduates.

“Slop mathematics”

The letter’s main complaint is how AI companies announce results. In the signatories’ account, a lab says its model has solved an outstanding theoretical problem, and the work of checking the proof and putting it to use falls to academic researchers. They call this “slop mathematics.”

After the latest result is dropped, research mathematicians are compelled to step in to properly verify, disseminate, and sometimes discredit entirely the claimed results. This labor goes uncompensated, uncredited, and unacknowledged.

The authors went further: “To put it bluntly, AI companies are engaging in research misconduct.” They objected to the mathathon on two grounds: using AI on open problems in the first place, and putting the companies whose research practices they were criticizing in the sponsor slot. They said the event would serve as unfair advertising for AI companies, which “will take credit for the effort of talented undergrads.”

The week before the letter

The letter landed at the end of a week in which OpenAI’s math claims were already under fire. On September 8, OpenAI said its agents had solved the Navier-Stokes equations, the formulas that predict how liquids and gases move and one of the Millennium Prize Problems. The company said the work ran on an internal model “significantly more capable” than its latest public ones.

Tristan Buckmaster, an NYU mathematician who had published research on the same problem with another academic the day before, said on Tuesday that OpenAI might have used his conversations with AI chatbots, OpenAI’s among them, to reach its solution. OpenAI did not deny it. It said only that it couldn’t rule out that “de-identified data derived from their usage of our products helped improve our models.”

Buckmaster was not the only one. Andreas Thom, a geometry professor at TU Dresden, wrote on Mastodon that his own chats with ChatGPT may have contributed to the results OpenAI announced, and accused the company of dishonesty about its training data. His accusation was reported on September 10, the same day OpenAI left the mathathon.

In that context, the hackathon seems to have become the nearest place where mathematicians could push back on how labs take credit for results. The letter’s two complaints, unpaid verification work and undergraduates’ effort used as advertising, are the same argument over credit that Buckmaster and Thom were making, applied to a student event.

What the organizers did next

The student organizers answered in a letter of their own on Thursday, saying they had not meant to cause such a stir. “Events like Mathathon encourage young people to stay excited about mathematics, engage their interests with modern tools, and maintain hope in a confusing time,” they wrote. “It’s our first time running an event at this scale, and we are grateful for all the feedback we have received.”

The open letter kept gathering signatures. EdSource later counted 2,038 academics who had signed it. The same outlet reported that the organizers then redesigned the November event. Teams will no longer try to solve open problems. Instead they will write alternative proofs for problems that already have solutions mathematicians consider unsatisfying, and the event will not be sponsored by companies that build proprietary AI models.

“We came to the decision that we want to put up an event that unites the math community, rather than divides it,” Zhao said.