Robust Cooperation in the Prisoner's Dilemma — AI Alignment Forum