When I asked ChatGPT about a paper written by a co-author and I, it made up some very plausible (and almost correct) material, the screenshot just shows a portion. However, because I was using the AI in my area of expertise, I could spot the issue immediately. Asking the AI to report when it is confident helps, too.
Being clear, giving the AI examples, and giving it step-by-step directions makes a big difference.
