I don't think I've ever encountered something written by an LLM that was better than barely passable prose. It's just about suitable for documentation and mundane news regurgitation, but a piece of creative writing? Please show me your favourite examples where it's better than "passable", I'm open to changing my mind.

Figuring out whether something is written by AI or a human is fairly easy with Pangram, and soon all major models will embed fingerprinting and it will be basically 100% detectable.

A personal account doesn't change the fact that AI will become so ubiquitous that most people won't be able to tell for sure whether something was written by a model, let alone a model following specific instructions like "Write like Mark Twain."

That's a consequence of not using enough compute. The correct way to have an LLM write creatively is to have it try hundreds of permutations and then judge those hundreds. This is basically what humans do anyway. Some models are better than others but you can absolutely write well with it. I generally find that when people say it doesn't work for them they aren't actually trying very hard or they simply do not know what they are doing.