>My only caution is that good writers and LLMs look very similar, because LLMs were trained on a corpus of good writers.

While that might be ideal, is that really the case with most LLM training data? Does the curation process weed out all the slop from bad writers?