OpenAI Challenged by 25 Fields Medalists and Withdrawal from CalTech

Mathematicians fear that the use of LLMs may fuel haste, complicate attribution, and push for secrecy. A proof attributed to OpenAI is said to be unverified, and the company has withdrawn sponsorship from CalTech. A professor from NYU further accuses it of attempting to downplay a credit, while 25 Fields Medalists call for clear rules.

Fear of Secrecy, Disputed Attribution, and Questionable Validation

Leading laboratories may invest tens of millions of dollars in the use of LLMs to try to outpace researchers behind a proof, which, according to the open letter, fosters a culture of secrecy. The signatories criticize the speed at which certain solutions are announced, leaving little time for thorough writing, identification of new methods, and citation of prior work. The proof attributed to OpenAI is reported to remain unverified. Significant risks regarding attribution and plagiarism, as well as broader concerns for the preservation of open research culture, are raised. Some mathematicians question whether their use of Codex has been integrated into new models from OpenAI. According to the signatories, the value of AI-generated solutions will only materialize if the mathematical community can understand and communicate them beyond its circle.

Individual Accusations and Withdrawal of Sponsorship from CalTech

Tristan Buckmaster, a professor at NYU, claims that OpenAI pressured him not to credit a collaborator employed by Anthropic in solving an important mathematical problem. He also questions whether OpenAI used their work with Codex to produce its own proof during an intensive inference weekend. Furthermore, on Thursday, OpenAI withdrew its sponsorship from a mathematical event at CalTech after being criticized by researchers from the university.

A Structured Mobilization Around 25 Fields Medalists and Leiden

Twenty-five Fields Medal winners have signed an open letter to raise awareness about the threat that AI laboratories pose to their intellectual work. The signatories believe that without the engagement of mathematicians to develop and integrate ideas stemming from AI into the mathematical corpus, human transmission among researchers would be compromised. Their initiative follows the Leiden Declaration, published in June by a working group, which analyzes the impact of proofs produced by LLMs on mathematical practice and formulates recommendations for mathematicians, institutions, and policymakers. The importance of an intellectual superstructure that trains students, raises new questions and ideas, and integrates these advancements into general culture is also advocated. They draw parallels between current difficulties and those faced in other scientific and creative professions, emphasizing that this is an issue for all of humanity: ensuring that the evolution of work methods under the influence of AI does not lose sight of the original objectives of these endeavors.