OpenAI announces new scientific results in mathematics and computer science and delays publication of physics materials for review

OpenAI announced a new set of scientific results reached by its models in mathematics, computer science and physics, after weeks of posting scientific manuscripts exceeding 700 pages that included solutions and progress on 372 major mathematical problems.

The company took a different approach this time in handling what its models produced, publishing the mathematical discoveries to enable researchers to study and interpret them independently, while enlisting external expert groups to examine and explain the computer science and physics results.

In computer science, OpenAI asked the expert group to prepare papers and scientific materials explaining the results produced by its models, while making the AI outputs available to the public, according to press reports.

As for the physics-related results, the company postponed publishing their materials until the group assigned to examine them completed its work, according to press reports, and the group set up an email address to receive inquiries about its work: YANGMILLS4D[@]gmail.com.

OpenAI added two new scientific manuscripts to GitHub and made them available to researchers. The first manuscript addresses what was described as a proof of a principle known as Sogge's Local Smoothing, concerning the behavior of solutions to the wave equation across different spatial dimensions.

The second manuscript presents a proof claiming that the Exponential Time Hypothesis (ETH) is false. This hypothesis is an important assumption in theoretical computer science, suggesting that some computationally hard problems cannot be solved by fast algorithms within certain time bounds.

ETH enjoys wide acceptance among researchers and has formed the basis for hundreds of papers in areas such as databases, computational biology and engineering. If the hypothesis were proven false, that could open the door to developing more efficient algorithms for hard problems, according to the manuscript.

Notably, the scientists who proposed the ETH are currently participating in the expert group OpenAI hired to examine and explain the results, according to reports.

In physics, the results concern proofs related to interacting quantum field theories that include gauge fields and fermions in four dimensions. The group tasked with examining these results has not yet published scientific materials explaining them, according to the reports.

OpenAI said it will not seek academic recognition or scientific credit for the discoveries made by its models, noting that members of the research groups will receive credit for the papers and materials they prepare to explain the results, and that these groups will ensure proper attribution to researchers and to the methods the models drew on.

Publishing the results and subjecting them to specialist review remains essential to determine their validity and significance, and whether they will lead to revisions of prevailing assumptions or open new avenues for scientific research.