OpenAI reports AI-generated results in math, computer science and physics
The company has published two manuscripts on GitHub and enlisted outside experts to examine and explain its computer science and physics findings.
OpenAI has announced new results generated by its models across mathematics, computer science and physics, including a claimed disproof of the Exponential Time Hypothesis. The company is using different approaches to publish and explain the findings, with outside expert groups involved in two of the fields.
Two manuscripts are available on GitHub. According to OpenAI, one establishes Sogge's local smoothing for waves across all spatial dimensions, addressing a conjecture concerning wave-equation solutions. The other presents what the company describes as proof that the Exponential Time Hypothesis is false.
That hypothesis holds that certain computational problems cannot be solved quickly. OpenAI said hundreds of papers in fields including databases, computational biology and geometry have relied on it. The scientists who originally proposed the hypothesis are among the experts asked to examine and explain the models' results.
OpenAI has made the mathematics findings available for researchers to investigate independently. For computer science, it has also published the AI output and asked external collaborators to prepare their own explanations. The physics group will handle publication of its materials but has not yet published anything about the findings, which OpenAI says concern interacting four-dimensional quantum field theories with gauge fields and fermions.
The company said it would forgo academic credit for the discoveries. Collaborators will receive credit for the materials they produce and will be responsible for appropriately acknowledging the researchers, methods and ideas used by the models.
