Scott Aaronson on "Reform AI Alignment" — LessWrong