This is part of the work that lead to Aristotle, the system that performed at Gold level at IMO: https://arxiv.org/abs/2510.01346
Very interesting. Do I get this right, running 500000 instances for 1 hour can be done for about $5000, or are there many hidden costs? (500000 * $0.01).
Lean4 with a mathlib project seems really slow has anyone else experienced that?
am i understanding it right that this is used to validate the output of llms? any other uses for distributed lean? genuinely curious
Lean is an automated theorem prover. It decides if a given proof is true or not. This uses LLMs to try to write proofs for a given problem