Alignment as Distribution Learning: Your Preference Model is Explicitly a Language Model (2506.01523v1)

Abstract: Alignment via reinforcement learning from human feedback (RLHF) has become the dominant paradigm for controlling the quality of outputs from LLMs. However, when viewed as `loss + regularization,' the standard RLHF objective lacks theoretical justification and incentivizes degenerate, deterministic solutions, an issue that variants such as Direct Policy Optimization (DPO) also inherit. In this paper, we rethink alignment by framing it as \emph{distribution learning} from pairwise preference feedback by explicitly modeling how information about the target LLM bleeds through the preference data. This explicit modeling leads us to propose three principled learning objectives: preference maximum likelihood estimation, preference distillation, and reverse KL minimization. We theoretically show that all three approaches enjoy strong non-asymptotic $O(1/n)$ convergence to the target LLM, naturally avoiding degeneracy and reward overfitting. Finally, we empirically demonstrate that our distribution learning framework, especially preference distillation, consistently outperforms or matches the performances of RLHF and DPO across various tasks and models.