Whenever someone attending the tech meetup I frequent gets a little too jazzed and excited about AI, I pull out my phone, open Emacs (I use termux, btw), bring in a corpus of text (an excerpt from A Christmas Carol is good), and say M-x dissociated-press.

Then I say "What this is doing is predicting the next character based on statistical likelihood of the previous few characters based on thencorpus of text. And fundamentally, that's all ChatGPT does -- predicting the next symbol based on a statistical model. ChatGPT has a much more sophisticated statistical model than this simple Markov chain and a vastly larger corpus, but really it's just doing the same thing." And we have a giggle about the nonsense DP makes of Dickens, but then I say that ChatGPT emits nonsense too, but it's far more insidious nonsense because it is much more plausible sounding.