OpenAI denies using mathematician's unpublished research without permission: Codex draft had no impact on Navier-Stokes proof.

Beating AI News Flash: OpenAI has issued its latest response to the Navier-Stokes credit-grabbing controversy, explicitly denying that it stole unpublished research from NYU mathematician Tristan Buckmaster. After an internal investigation, the company stated that the prompts Buckmaster submitted to Codex over the past two months could not have influenced this internal model in any way, including through training.

Buckmaster had previously been researching the same direction as Levent Alpöge, and a large number of drafts were placed into Codex. After their progress reached OpenAI, OpenAI concentrated computing power and produced a Navier-Stokes proof within days. Buckmaster therefore suspected that his unpublished research might have already been learned by the model. OpenAI has now explicitly denied this point.

OpenAI also drew the temporal boundary at July 3, saying that any user input after that date could not have affected this system. Some people in the community speculated that this date may correspond to a certain training data cutoff or model checkpoint, but OpenAI did not explain.

Buckmaster and Alpöge's key breakthrough occurred in mid-August, far later than the July 3 boundary drawn by OpenAI.