The hard part about working with AI believers is trying to come up with prompts that give outputs that don't look totally stupid so that I can say "yeah, here's the process you wanted me to AI-ify". I think this may be related to me reading fairly fast, that I've yet to have a "summary" or analysis of text come out of an LLM that was generated faster than I read, and in any way a meaningful representation of what I think the salient points of the document were.