chatgpt is definitely capable of saying it doesn't know, but that itself is just guessing at its own capabilities because a prompt has primed it with heuristics for what kinds of things it shouldn't try to produce
because the underlying llm doesn't even have a concept of knowledge