Re-formalizing PD — AI Alignment Forum