Towards Robust Alignment of Language Models: Distributionally Robustifying Direct Preference Optimization | Read Paper on Bytez