Yes, this is what I mean: the thing the LLM is doing is a statistical guesswork process - that's the core of what the thing is. In many cases, its statistical guesswork is correct - it's guessed the right thing. That is not the same thing as actually referencing a canonical source, and should not be confused as such. So long as people's mental models of LLMs don't match what the LLM is actually doing, we'll continue to be surprised by how often they're wrong about things and we'll continue to try to use them in ways we shouldn't. Recognize what the tool you're using is and use it appropriately.