Hi,
Probably some on this list already checked how well chatgpt answers
questions about domains one knows well and then probably noted that in
spite of impressive wording, one can run into quite incorrect answers.
One can get really stupid responses about tex and friends, but also can
get impressive exmapled when asked for. (I'm still planning a wrap up of
some.)
That said, one should read:
https://cs.stanford.edu/~knuth/chatGPT20.txt
and this makes a pretty nice new sample file:
It's amazing how the confident tone lends credibility to all of that
made-up nonsense. Almost impossible for anybody without knowledge
of the book to believe that those "facts" aren't authorititative
and well researched.
(among some other remarks)
I wonder what Knuth would say about having ChatGPT write computer programs, which I think can be outright dangerous. Either you get shitty code which doesn't work, which is good, or you get shitty code which works which is bad or Really Bad.