AI Safety Thursdays: Understanding The Self-Other Overlap Approach — LessWrong