Information-Theoretic Reward Decomposition for Generalizable RLHF

Re-assign community