The self-fooling problem. — LessWrong