OpenAI exits Caltech math hackathon over criticism

OpenAI has officially pulled out of sponsoring Caltech’s upcoming mathematics hackathon. The decision came after a group of current and former mathematicians at the California Institute of Technology issued an open letter criticizing the event. They argued that AI-sponsored competitions create a harmful dynamic where large language models generate unverified claims that human researchers are then forced to fix, verify, or debunk without compensation or credit.

The controversy intensified earlier this week when OpenAI announced that its agents had solved the Navier-Stokes equations, a set of formulas used to predict fluid motion that has remained unsolved for nearly a century. However, Tristan Buckmaster, a mathematician at New York University, suggested that the AI’s solution may have relied on his own previous conversations with chatbots. OpenAI acknowledged that it could not rule out the possibility that de-identified data from user interactions helped improve its models, a point that fueled accusations of intellectual theft.

Critics label AI output as mathematical slop

The open letter, cited by GN technics/ai (en-US), coined the term "slop mathematics" to describe this phenomenon. The academics explained that AI companies often claim to have solved outstanding theoretical problems, but they leave the hard work of proper verification, dissemination, and sometimes total discreditation of those results to actual researchers. This labor, they noted, is uncompensated and unacknowledged, placing a significant burden on the mathematical community.

Dan Roberts, OpenAI’s research lead, responded to the criticism by acknowledging the disruptive nature of rapid AI progress in mathematics. He stated that the company is now looking for better ways to engage with the math community to integrate this technology responsibly and communicate its impacts. However, the trust deficit remains significant, as many academics feel the current model of AI involvement prioritizes corporate branding over scientific integrity.

Event organizers deny intent to cause harm

The hackathon, known as Mathathon, was designed as a practice ground for young mathematicians to use AI tools responsibly to augment their understanding. Organizers, many of whom are undergraduates, stated in a response letter that they did not intend to cause such a stir. They emphasized that the event was meant to keep young people excited about mathematics and help them engage with modern tools during a confusing time for the discipline.

The competition structure involved a first round where participants had 40 hours and $20,000 worth of AI tokens to solve an open problem. Teams with promising, well-explained answers would advance to a second round, receiving six months and additional tokens to further their work. Anthropic, another sponsor, donated a portion of a total $2 million in AI credits, though it remains unclear how much of that sum came from OpenAI or what happens to those credits now that the company has exited.

Unresolved questions about data usage remain

The core issue for many mathematicians is the opacity of how AI models generate solutions. If AI systems are trained on or influenced by the private research conversations of individual academics, the line between tool and collaborator becomes blurred. This creates a trade-off where companies gain proprietary improvements to their models while the human experts who provided the underlying intellectual labor receive no attribution. The withdrawal of OpenAI highlights the growing tension between rapid technological advancement and the established norms of academic credit and verification.