LawZero: Safety from Honesty in a Disinterested AI Predictor · HackerLangs