AGI safety from first principles: Alignment — AI Alignment Forum