AI & ML · Sep 10
OpenAI's Navier-Stokes files make the proof checkable, not just callable
The real lesson is the move from press call claims to Lean artifacts, because science needs receipts and theorem provers do not care about vibes.
- Treat AI science claims as artifact problems: look for proof files, checks, and independent review paths.
- Use verification loops in your own AI workflows, especially when outputs are complex or high stakes.
- Separate correctness from provenance: a valid artifact still needs transparent credit, data, and process records.