Mathematicians split over OpenAI's proofs and security implications
OpenAI's release of mathematical solutions has fractured the academic community, with some praising the utility and others calling for a halt to collaboration. The absence of cryptography proofs in the releases has raised concerns about encryption security.
OpenAI released mathematical results narrowing the solution to the chromatic number of the plane problem to either six or seven colors and solved Paul Erdős's unit distance problem. The company also claims to have solved the Navier-Stokes Millennium Problem using 10,000 AI agents over 88 hours. A controversy arose regarding the Navier-Stokes proof, as NYU mathematician Tristan Buckmaster claimed he was working with an Anthropic researcher near the breakthrough and had saved his work in an OpenAI program, raising questions about whether the AI or human work was the source. No human currently fully understands the Navier-Stokes proof, and errors have been found in some other AI-generated proofs.
The mathematical community has fractured into distinct camps regarding AI adoption. Terence Tao, an optimist, stated that models are ready for real work, saving more time than they cost, and envisions an "industrial mathematics" with human-machine teams. Conversely, 25 Field Medalists signed a statement arguing that AI company goals are misaligned with mathematics, a position supported by 7,400 researchers. The Association for Human Mathematics, formed in response to OpenAI's recent releases, called for a halt to collaboration with OpenAI, viewing the developments as a power demonstration. Skeptics like Stephen Wolfram downplayed the utility of the discoveries, while another mathematician, Goers, argued mathematics would die from oversupply rather than stagnation. Tao also noted a "proof digestion" problem, questioning the value of mathematics if proofs cannot be verified. Observers note that proofs related to encryption and cryptography are notably absent from recent AI releases, suggesting a deliberate withholding to prevent immediate security breaches and implying that AI models may have already compromised current encryption standards.