x
Alignment through atomic agents — LessWrong