My go-to topic to test it with is talking about characters in the movie Hackers. I noticed in someone else’s session that it would take their correction but still hang on to the incorrect contradictory belief. In my session it came up with a seeming rationalization. So I tested it just now, trying to provide the correct information and directly contradicting the incorrect information (x is A. x is not B). That helped, but eventually it just choked. It seems to be handwaving and guessing (bluffing, bullshitting) at the most likely and generic answers when it doesn’t know something.

I’m thinking ChatGPT is best used for generating ideas, not factual information.