An Epistemological Nightmare — LessWrong