Nobody expects the image generators to produce images of real life. The LLMs are doing the same kind of thing with words. I'm at a loss as to how people who presumably understand how it works to expect it to be factually correct. It's not a knowledge engine. Some have described it as a reasoning engine.
In other words, being right 99% of the time is a really bad idea to base your next system on
That's where the bar is right now. Literal omniscience and infallible foreknowledge.