Provable Zero-Shot Generalization in Offline Reinforcement Learning | Read Paper on Bytez