Webinar: First Proof, Third Batch Results
Ten research-level problems, solutions from several AI systems, refereed blind by experts. Wednesday, October 14 at 3:00 PM ET, with live Q&A.
Wednesday, October 14, 2026 · 3:00 PM ET · Online via Zoom
Can AI systems solve the kinds of problems mathematicians meet in their own research? First Proof puts that question to a public test.
For each batch, working mathematicians write ten questions drawn from their own research. Several AI systems attempt them, and experts in each problem's field grade the AI-generated solutions blind. The systems' results are then compared, and the referee reports are published openly.
The third batch was graded on October 8–9 at ICARM in a two-day meeting. The solutions were assessed based on mathematical correctness, quality of exposition, and quality of attributions. The results will be announced in a webinar on Wednesday, October 14th at 3 PM EST.
Program
- Welcome from ICARM Director Jeremy Avigad.
- Overview of the results by First Proof editor Mohammed Abouzaid (Stanford).
- A problem writer's view: one of the third batch's problem writers gives a short presentation on their problem.
- Q&A with all four First Proof editors: Mohammed Abouzaid (Stanford), Nikhil Srivastava (UC Berkeley), Rachel Ward (UT Austin), and Lauren Williams (Harvard).
The webinar is free and open to everyone. Registration is required: register on Zoom.
To keep the conversation going after the webinar, join the #first proof channel on ICARM's public Zulip.