A thing I've noticed about GPT-generated text, something that stands out to me before I even notice inaccuracies or plagiarism:
It's unreadable. The writing style screams "skim me." I have seen GPTed paragraphs that were factually correct, and some that were even impressive in a dancing-bear kind of way, but I have never, not once, seen one that was a pleasure to read. There's something deeply numbing about a writer who always makes the most obvious choices.
I was thinking about this with that Gizmodo bot article on Star Wars. A lot of people pointed out that the article was wrong, but it's also deadly boring. It might make a kind of unethical sense as SEO stuffing, but as an article you expect people to sit down and enjoy? It reads like a TV Guide listing. It has no perspective or insight on the Star Wars series to share, not even a banal one. Just... these are some movies that exist.
LLMs are getting okayish at writing things that fulfill requirements, but they're nowhere near being able to write things that are meant to be read.
