[flagged]
This simply isn't true. An LLM can do a weak parody of a sufficiently-famous author, but they're extremely poor at sustained fiction writing even without trying to emulate a specific style.
Perfect grammar is only a small part of writing well.
You see a lot of crap for the same reason most code is crap - the human who did the thing sucked. Language models aren't magic, they're just a compiler that guarantees an output regardless of input quality and structure.
That doesn't mean much wrt to the capability of the model.
The world is filed with crap literature, movies, stories, and advertisements written by honest to God human professionals in their respective field, and we trained these things in that environment. It's not surprising to me that they need a nudge to get to Hemingway when the corpus average is presumably so low.