I'd say that GPT-3 could probably beat a lot of the English-speaking (human) population at producing these though. I've read far worse on social media, that I've attributed to humans authors.
No doubt. Since it’s built out of a corpus of human produced text, it should be capable of producing human level text (but probably not often exceeding it). Yet many people are not capable of writing syntactically correct text.
As well as being syntactically correct, it also reflects the contextual patterns of the tokens used to train it and then prompt it, which is perhaps limited by both intelligence and knowledge in humans.
