It's weird to me that people are still thinking of AI as something that can be taught, when of course an autocomplete machine doesn't understand facts.

If you specify a worldview and a context in the prompt, a lot of this can be avoided, but there don't seem to be any real failsafes.