@simon that would require you to have a fairly great NLP system to detect asserted facts and so on because LLMs aren't more or less "sure" about any specific long sections of text, generally. They sample from a distribution of probable continuations of some specific input text.