Second Batch Community Experiment
The community experiment is an informal activity connected to the second batch. It gives mathematicians and interested participants a separate space to test AI systems on selected research mathematics problems, compare transcripts, and discuss what the attempts reveal. The community experiment problems have not appeared on the internet or in any public forum, but some of them may be parts of papers under review. They have solutions of at most 9 pages which are known to the authors.
Guidelines & Suggestions
Please preserve a complete transcript of your process, including models used, code, number of attempts, human interaction, token use, cost, and wall clock time.
Please do not post solutions publicly until the discussion period begins.
Here are some possible axes for experimentation.
- How cheaply can you produce a solution?
- What is the smallest model you can use to produce a solution?
Week #1
Problem release
A separate set of problems is made available for public experimentation after the formal benchmark.
Open experimentation
Participants try the problems with AI systems and are encouraged to preserve complete transcripts.
Solutions posted
Solutions are released so attempts can be compared against complete proofs.
Discussion
The community can compare methods, failure modes, and evidence about mathematical capability.