3 ms·
LLM-generated skills work, if you generate them afterwards
- verdverm 7mo agoI generally think this is a valid point to make, but there is some complexity in how you generate input to benchmark problems after the fact. You've essentially given it the answer, based on the results of a human-ai interaction