Abstract
Learning from human feedback (LHF)—and in particular learning from pairwise preferences—has recently become a crucial ingredient in training large language models (LLMs), and has been the subject of much research. Most recent works frame it as a reinforcement learning problem, where a reward function is learned from pairwise preference data and the LLM is treated as a policy which is adapted to maximize the rewards, often under additional regularization constraints. We propose an alternative interpretation which centers on the generative process for pairwise preferences and treats LHF as a density estimation problem. We provide theoretical and empirical results showing that for a family of generative processes defined via preference behavior distribution equations, training a reward function on pairwise preferences effectively models an annotator's implicit preference distribution. Finally, we discuss and present findings on "annotator misspecification"—failure cases where wrong modeling assumptions are made about annotator behavior, resulting in poorly-adapted models—suggesting that approaches that learn from pairwise human preferences could have trouble learning from a population of annotators with diverse viewpoints.
Authors
Vincent Dumoulin, Daniel Johnson, Pablo Castro Rivadeneira, Hugo Larochelle, Yann Dauphin
Venue
Transactions on Machine Learning Research (TMLR)