← Back to context

Comment by baq

8 hours ago

Dropping ultra hard problems on those things and giving them reward when lean accepts the proof sound like an obvious RL environment tbh