The Navier–Stokes Millennium Prize Problem

The Navier–Stokes Millennium Prize Problem, one of seven challenges with a $1 million reward since 2000, has been claimed to be solved by OpenAI's model after using GPT-5.6 Sol. However, the claim has been met with accusations of skulduggery from Tristan Buckmaster and Levent Alpöge, who were working on related problems. The OpenAI team claims they heard about the resolution through rumors and reached out to Tristan and Levent, but they were not invited as co-authors.

The OpenAI agents spent about 88 hours evaluating the result on all open Millennium Prize problems, sending 2.7 million messages and using 130 billion output tokens. OpenAI's approach to this problem mirrors concerns in the world of computer security, where rumors of unpatched vulnerabilities can lead to the discovery of exploits and potentially millions of dollars in spending to address them first.

This raises questions about how data used to improve model performance is actually defined and used.