The problem exist, the conclusion is wrong. LLMs DON'T produce great content, they PRETEND to produce great content, even in coding. They're as good as the statistical incidence. Writing a book with an LLMs, like with code, could lead to good results sure, but they're in the mean, sometimes slightly less sometimes slightly more. When it's good it looks good, because that's how it works, and it's broken way more subtly than ever