"Knowing what they don't know" is one of the big unsolved problems in llm training. There's a finite number of facts. There's an infinite number of non-facts.

"Knowing what they don't know" is one of the big unsolved problems in llm training. There's a finite number of facts. There's an infinite number of non-facts. If it could do what you're suggesting, it wouldn't need to do it in the first place.