Comparing Four Approaches to Inner Alignment — LessWrong