OpenAI Expands AI-Driven Discoveries Across Mathematics, Computer Science, and Physics

To analyze and interpret these findings, OpenAI assembled teams of subject-matter experts.

Samuel Boivin/Shutterstock

Shortly after publishing more than 700 pages of solutions and advancements covering 372 significant mathematical problems, OpenAI has disclosed additional breakthroughs. The latest findings extend beyond mathematics to include results in computer science and physics.

For this release, the company tested three distinct approaches. OpenAI has made the mathematical results publicly available so researchers can review them and develop their own explanations. For the computer science and physics findings, however, it assembled external panels of expert collaborators. The group assigned to computer science was asked to produce independent writeups and supporting materials, while OpenAI simultaneously released the underlying AI-generated output. Regarding the physics results, publication has been deferred to the relevant expert group. OpenAI is inviting feedback on how to expand participation and improve the usefulness of this process.

Two new manuscripts are now available on GitHub. The first demonstrates Sogge’s local smoothing for waves across all spatial dimensions, addressing a conjecture concerning the behavior of solutions to the wave equation. The second presents evidence that the Exponential Time Hypothesis—a computer science assumption stating that certain problems are inherently too difficult to solve quickly—is false.

OpenAI points out that the Exponential Time Hypothesis is broadly accepted and has underpinned hundreds of papers across fields such as databases, computational biology, and geometry. If the hypothesis is incorrect, researchers could theoretically develop more efficient algorithms for solving problems in these areas. The scientists who originally proposed the hypothesis are now part of the group tasked by OpenAI with examining and explaining its models’ results.

The physics findings concern a proof involving interacting quantum field theories with gauge fields and fermions in four dimensions. The group reviewing the physics results has not yet published any material, but has provided an email address—YANGMILLS4D@gmail.com—for inquiries about its work.

OpenAI stated that it will not claim academic credit for discoveries made by its models. Instead, members of each group will receive credit for the writeups and materials they produce. These groups will also be responsible for ensuring that AI-generated results properly acknowledge the researchers, techniques, and ideas on which the models relied.

Source link