Could anyone comment on how often Dafny finds proofs for useful theorems on its own in practice, without requiring the programmer to use expert knowledge of the proof search system to break it down into a sequence of gettable lemmas and/or refactor the code?
If the answer is "not that often", is there a possibility that LLM integration could help with proof search? This seems like an application where getting it 95% right is actually OK since proofs generated by the LLM can be automatically checked for correctness.
If the answer is "not that often", is there a possibility that LLM integration could help with proof search? This seems like an application where getting it 95% right is actually OK since proofs generated by the LLM can be automatically checked for correctness.