Improving LLM General Preference Alignment via Optimistic Online Mirror Descent

Open in new window