Personally, I could probably hand write a book of paragraphs, all very distinct and different to each other, you likely wouldn't even know they were by the same author if you didn't use stylometry ML, and you'd probably have a rough time if I used adversarial stylometry to counter that.
You'd think by this point that music would have all been done too, but people keep finding novel stuff in 12TET, 4/4 and sound design. That's without even getting into other time signatures, scales and microtonality.
Can't you specifically ask LLM to tailor the paragraph to produce those variations that you will incorporate to make your own writing different each time? Most of the criticism of LLM not being able to produce something is just people asking LLM to basically MIND-READ the exact thing they had in mind.
Well, can't anybody else just tell it the same prompt? The output is generic and samey between all users because the average human being enters the average prompt, thus getting the average output from the training data of statistical averages because it's a next-word prediction model.
If I run a writing workshop and tell a room of a hundred people to write me a story about two star-crossed lovers from different time periods, the likelihood of them writing the same thing in the same way is extremely low statistically speaking. If the week after I then give that same room of people the same exercise, they will write something different again, even if the core story that's being told is the same thing. That's because human brains aren't static, and they don't store information as binary values, the "data" is abstract and chemically gradient. That's without even accounting for mood, the general zeitgeist, and what they've been through since then, and not even getting into them entertaining themselves by changing it up or learning from the previous one.
Long-form consistency also isn't something LLMs do well with.
You'd think by this point that music would have all been done too, but people keep finding novel stuff in 12TET, 4/4 and sound design. That's without even getting into other time signatures, scales and microtonality.