Simple question about corrigibility and values in AI. — LessWrong