AI Alignment and Recognition — LessWrong