Is there something about an LLM that makes it incapable of simplifying a codebase? I have only seen that the slop jockeys who wield LLMs most ardently don't seem to care about that, and that older models were worse at everything, including that.