OpenAI Announces New Discoveries In Mathematics, Computer Science And Physics
OpenAI has brought together groups of experts to study its new discoveries and explain the results.
Just a few weeks after OpenAI released papers with more than 700 pages of solutions and progress on 372 major math problems, the company has announced more discoveries. This time, the results cover not only mathematics but also computer science and physics.
For this release, OpenAI tried three different approaches. It published its math results so researchers could read them and try to explain how they work. For the computer science and physics results, it brought in groups of outside experts. OpenAI asked the computer science experts to explain their results in papers and other materials they wrote themselves. It also made the AI's results available to the public. For physics, OpenAI has left the decision to publish the materials to the group studying them. The company welcomes feedback "on how to broaden participation and make this process more useful."
OpenAI has added two new papers on GitHub that people can read now. The first proves Sogge's local smoothing conjecture for waves in every number of spatial dimensions. This conjecture is about how solutions to the wave equation behave. The second paper presents a proof that the Exponential Time Hypothesis (ETH) is false. ETH is an idea in computer science that says certain problems cannot be solved quickly.
OpenAI says that many scientists believe ETH is true. Hundreds of research papers across many fields, including databases, computational biology and geometry, have relied on it. If ETH is actually false, it might be possible to create better algorithms that solve problems in these fields more efficiently. The scientists who first proposed ETH are now part of the group OpenAI has asked to study and explain the results produced by its AI models.
The physics results include proofs about interacting quantum field theories with gauge fields and fermions in four dimensions. The group studying these results has not yet published anything about its work. However, it has shared an email address, YANGMILLS4D[@]gmail.com, for people who have questions about its efforts.
OpenAI says it will not claim academic credit for discoveries made by its AI models. Instead, the people in each group will receive credit for the papers and other materials they create. The groups will also make sure the AI-generated results properly acknowledge the researchers, methods and ideas that the company's models drew on.