Comment by threatofrain

I've seen a different kind of usage pattern recently, which is to find a problem that the LLM is good at, something which is very local in reasoning, and do it at big scale across the whole codebase.