Formalizing Newcomb's — AI Alignment Forum