Only when it adds value and still sounds human.

I avoid obviously AI-written prose. If it wasn’t important enough for the author to put effort into writing it, why should I put effort into reading it?

Only when it answers a question i asked.

A 70 page summary a human expert in that field does not reject is fine, you do see AI prose padding in there.

If it's good enough, sure. Not much of it is.

But there's also the whole epistemic issue of "if it's good enough, you won't know it's AI." My AI-dar is pretty keen but given the quantity of things I read and the quantity of AI stuff out there, it's entirely likely something I thought was human and liked was actually AI. In which case, there doesn't seem much point in getting retroactively upset about it if someone later tells me. shrug

Enjoy? No, but a lot of reading and writing is very functional, for communication purposes for business etc. (documentation is a good example). In these cases the important thing is that it's accurate.

Sometimes i realize I'm reading AI bullshit while I'm already a few paragraphs deep (and then auto-close). I get this feeling that something is "off", don't know how to explain. This happens more often with topics like movies and gaming news. I guess it is just easier to fake that kind of text to a perceived sufficiently quality threshold. Maybe because most of the human-written text in the genera already looks LLM-ish (repetition, conjecture, listicles...)

When it answers something I asked, or when I play with a small model with llama.cpp, I sometimes find their reasoning funny while they go in circles.

Yes, when it adds value and informative, specially when it summarizes the core idea of 100 pages

Yes when I plop down into an unfamiliar codebase and ask for a synopsis and introduction.

When I ask for diagnostic of an error I don’t understand.

When I ask for a first draft of documentation (way easier to edit than generate from nothing)

When I ask how to do something obscure buried deep in a configuration setting