OpenAI announced it is testing three different release strategies for AI-generated claims in mathematics, computer science and physics, working with external expert groups to interpret the outputs.
The company said it asked external experts to act as shepherds for three separate results concerning harmonic analysis, complexity theory and quantum field theory. Each field uses a different approach: the mathematics result releases the AI output for researchers to examine independently, the computer science result forms an external collaboration while making the AI output publicly available, and the physics result defers material release decisions to an external group.
OpenAI characterized the AI outputs as “raw research data” requiring expert interpretation to become scientific understanding. The company said it will not seek academic credit for discoveries its models produce, and that inviting researchers to steward collaborations does not imply they made the original discovery.
The external collaborations operate under four stated principles: participants are not required to sign nondisclosure agreements and may discuss the work publicly; working groups receive recognition for explanations and materials they create; participation is voluntary and unpaid; and participants’ conclusions are independent.
For the computer science collaboration, OpenAI published a natural language proof and Lean formalization to its GitHub repository. The company said it welcomes feedback on which of the three approaches proves most useful.
https://openai.com/index/building-scientific-understanding-with-academic-communities