Not too different from the ones I write by hand. If someone writes a “clean proofs” book maybe we can all (humans and ais) learn how to write better lean proofs. The pre-scale answer was contribute to Mathlib and get mentorship through reviews, but that doesn’t scale.