P.S.: Lest anyone think I'm a Luddite--not that I think that appellation is actually that bad--an example where I would use an LLM would be to help me generate

P.S.: Lest anyone think I'm a Luddite--not that I think that appellation is actually that bad--an example where I would use an LLM would be to help me generate synonyms or alternate inputs to a more-traditional search for a discrete external piece of information. For example, I might remember a book with a jester playing a lute and singing about ogres, and I just can't find any clear search results, because it was actually a bard strumming a harp with a poem about giants. This is much less dangerous than just throwing every dang thing into LLM inputs (since prompt injection isn't a fluke, it's a way of life) or filtering the results back through the same model in an opaque fashion.