Yudkowsky on AGI ethics — LessWrong