Further or Alternatively Stuart Buck Nathan Labenz Gary Marcus Ezra Klein Yes, it seems the helpful way to think of ChatGPT is as a plausibility maximizer that remembers vast quantities of documents but learned mostly details and phrases (and can’t reason or calculate). You should trust it least when you’re least able to notice it’s wrong.

For those who are impressed with Michal Kosinski's claim that LLM's are developing a theory of mind: