Adversarial System Hats — LessWrong