Improving LLM General Preference Alignment via Optimistic Online Mirror Descent | Read Paper on Bytez