My hardware prof said "bad data is worse than no data, because it deceives us".
This is where I am with LLMs; why I still don't use em (outside of research). I can believe that the fourth beast is no worse than most of you lot. But I was already struggling to process you.
Follow
(Obviously there's lots of places where semi-fictitious data is fine, like idea generation, art, idle talk, and what we called "spraffing": the conversational annealing that comes to you when you lie horizontal on grass and just talk without logical or artistic intent)
(I still don't use it there for aesthetic reasons but maybe I should)