Adversarial epistemology — LessWrong