Comment by brandonb

What are the advantages of reinforcement learning over DPO (Direct Preference Optimization)? My understanding is that the DPO paper showed it was equivalent to RLHF, but simpler and more computationally efficient.