x
AI Alignment Requires Systems That Distrust Their Own Optimization — LessWrong