[joe_the_user]

It's definitely not accurate to view that sort of prediction error or other internal value with an overall measure of the confidence, accuracy, "truth" or etc of the language the LLM produces.