Tristan Buckmaster and Levent Alpege, a researcher from Anthropic, spent a year working on a proof for the Navier-Stokes equations, one of the seven Millennium Prize Problems. They utilized both Claude and ChatGPT via Codex to assist in their research. In early September, rumors emerged that Anthropic had solved the problem, prompting Buckmaster to notify OpenAI that they were nearing a publication.
OpenAI responded by claiming that their internal latest model had also solved the Navier-Stokes equations using an approach nearly identical to that of the researchers. OpenAI later admitted that their internal team began prompting the model to solve the problem only after hearing the rumors, just a few days prior.
Buckmaster questioned whether the new model had been trained on user sessions from Codex, suggesting the potential appropriation of their research. OpenAI declined to answer this question directly, stating only that the model did not access user chats during the actual solving process.
Buckmaster claims OpenAI offered him the chance to publish the the solution first and claim the $1 million prize, provided he removed Levent Alpege from the co-authorship. He further alleges that OpenAI representatives suggested he should not jeopardize his career if he refused the deal.