OpenAI Publishes More Than 700 AI-Generated Math Preprints, Sparking a “Slopocalypse” Debate

The research community is in an uproar after OpenAI published more than 700 mathematical preprints generated entirely by artificial intelligence on October 6. The San Francisco-based maker of ChatGPT posted the work on the software repository GitHub.

The release drew sharply divided reactions. Some mathematicians celebrated the apparent solution of a long-standing problem, while others criticized OpenAI for publishing hundreds of papers at once. Some described the moment as a “slopocalypse”, even if the mathematical content ultimately proves to be formally correct.

OpenAI’s mass release divides mathematicians

“The number of papers submitted simultaneously is staggering,” wrote Alvaro Lozano Robredo, a number theorist at the University of Connecticut in Storrs. “Why is OpenAI trying to solve hundreds of published math problems at once?” he asked on the social media platform X.

In an email to Nature, an OpenAI spokesperson said the company’s goal is to provide tools for mathematicians and scientists. “To accomplish this, we evaluate and develop models for difficult front-line problems in all of science,” the spokesperson wrote.

Levent Arposi, a mathematician at rival company Anthropic in San Francisco, said on X that the release could be a “sad story” for researchers whose projects may be usurped by OpenAI’s work. “But that should be put aside for today. This is clearly the most important moment in the history of mathematics,” he wrote, arguing that the release shows how quickly the technology has become useful to mathematicians.

Other mathematicians said that working with the mathematics community would have been a better approach. Ursula Martin, a mathematician at the University of Oxford in the United Kingdom, said that publishing preprints in this way was like “throwing a messy first draft at the community and expecting us to sort it out for them.”

“I don’t feel like spending time studying these preprints, so I’m not going to read these papers,” said Michael Harris, a number theorist at Columbia University in New York City.

What do the AI-generated math papers solve?

Most of the preprints did not contain answers to the remaining five Millennium Prize problems. The problems were selected in the early 21st century by the Clay Mathematics Institute, whose scientific headquarters are in Oxford, England.

An OpenAI preprint on the Navier–Stokes equations, published in September, addressed one of the original seven problems. Another of the seven was solved by mathematicians in the early 2000s.