Responsible Release of AI-Generated Mathematics
5 hours ago
- AI labs should stop testing advanced mathematical problems on proprietary models, prioritizing human understanding.
- For results that a human understands, follow traditional academic norms: preprint submission, peer review, and conference talks.
- For results not yet understood, labs must perform initial release steps: cite related work, improve exposition, deposit in open repositories, disclose model details, formalize proofs, and document usage.
- Labs must provide funding and support for activities that develop human understanding, such as conferences, workshops, and long-term research positions.
- Ensure broad, equitable access to AI models to prevent a two-tier system that alienates the mathematical community.