The maximum reward gain under KL-regularized LM alignment is a Jeffreys divergence term, estimable as covariance from base samples, with best-of-N approaching the theoretical limit.
Improving reinforcement learning from human feedback with efficient reward model ensemble
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 2years
2026 2verdicts
UNVERDICTED 2roles
background 1polarities
background 1representative citing papers
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.
citing papers explorer
-
Theoretical Limits of Language Model Alignment
The maximum reward gain under KL-regularized LM alignment is a Jeffreys divergence term, estimable as covariance from base samples, with best-of-N approaching the theoretical limit.
-
Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.