OpenAI acknowledged that data from two mathematicians’ use of its products may have helped improve its models
OpenAI acknowledged that data from two mathematicians’ use of its products may have helped improve its models
On September 8, OpenAI stated that its researchers and software agents had neither seen Levent Alpöge and Tristan Buckmaster’s work before publication nor accessed specific user data while solving the Navier-Stokes problem, a mathematical problem concerning the motion of a viscous fluid. The company also acknowledged the possibility that de-identified data from these mathematicians’ use of its products had helped improve its models. Alpöge and another participant in the discussions, Sébastien Bubeck, subsequently gave different accounts of their conversations about coordination.
The company distinguished between access to specific work and the origins of the model’s capabilities. For the team and agents, the issue was which material they had accessed while solving the problem; for the model, it was which data might have contributed to earlier improvements:
“Although it is unlikely, we cannot rule out that de-identified data from their use of our products helped improve our models.”
De-identified data is data from which identifying information has been removed. “Their” in the quotation refers to Alpöge and Buckmaster.
The services’ policies explain when content may be used to train a model. In its data policy, updated on March 13, OpenAI says that content from services for individuals, including ChatGPT, may be used for training unless the user opts out; temporary chats are excluded from training. For the API, which allows software to connect to the model, as well as ChatGPT Team and ChatGPT Enterprise, prompts and responses are excluded from training by default.
The dispute followed closely timed mathematical announcements. In its announcement of a Navier-Stokes solution, OpenAI said that on September 1 it had tasked agents with finding solutions to open problems and subsequently learned of Alpöge and Buckmaster’s result for the forced Euler equation, a model of fluid flow subject to an external force. OpenAI’s result concerns Navier-Stokes, so the results addressed different mathematical problems.
After publication, the participants publicly gave conflicting accounts of their coordination. Alpöge wrote that he had heard an offer of a Millennium Prize in exchange for being left out of the paper; Bubeck responded that he had not asked for Alpöge to be removed as an author of his own work.
For a research group, this raises two practical questions to address before working with a model: which service receives its materials for training, and how the parties document access to an unpublished result.