You are correct! I was fantasizing how a bad GPT answer could look like, similar to the errors that were in the answers to my prompts to GPT-3.5 for imagemagick

You are correct! I was fantasizing how a bad GPT answer could look like, similar to the errors that were in the answers to my prompts to GPT-3.5 for imagemagick. These prompts where precise and in good faith, with some rounds of clarification. I am sure newer GPT versions will do better, and I think for a tool as ubiquitous as git the answers from ChatGPT would have been perfectly fine too. Just wanted to illustrate why I wouldn't mske it a habit to ask it factual questions. The "lost two commits" answer was meant as whimsy joke. More realistic scenarios are more boring of course. E.g. memorizing wrong explanations for solutions that happen to work, or accepting wrong answers as fact when it's inconsequential for the short term acceptance of the answer. The same thing easily happens with Google's knowledge snippets they extract from search results.