Model Mis-specification and Inverse Reinforcement Learning — AI Alignment Forum