Paper of the DayResearch
Paper of the day: Direct Preference Optimization (DPO), explained
DPO showed language models can be aligned to human preferences without a separate reward model or reinforcement learning. How it works and why it spread so fast.
AI Desk3 min read