Pith. sign in

REVIEW 7 cited by

Correlated Proxies: A New Definition and Improved Mitigation for Reward Hacking

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2403.03185 v4 pith:3LPH36JO submitted 2024-03-05 cs.LG cs.AI

classification cs.LGcs.AI
keywords rewardhackingdefinitionproxypolicyrlhftrueacross
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

Because it is difficult to precisely specify complex objectives, reinforcement learning policies are often optimized using proxy reward functions that only approximate the true goal. However, optimizing proxy rewards frequently leads to reward hacking: the optimized reward function ceases to be a good proxy and the resulting policy performs poorly with respect to the unspecified true reward. Principled solutions to reward hacking have been impeded by the lack of a good definition for the problem. To address this gap, we introduce a definition of reward hacking based on the correlation between proxy and true rewards for states and actions seen by a "reference policy" that breaks down under optimization. We show that this definition captures reward hacking behavior across several realistic settings, including in reinforcement learning from human feedback (RLHF). Using our formulation, we show theoretically that regularization to the reference policy can effectively prevent reward hacking. While the current practice in RLHF applies a KL penalty between action distributions for this purpose, our theory suggests regularizing the $\chi^2$ divergence between the policies' occupancy measures can be more effective. We intuitively show the benefits of this type of regularization and demonstrate that it better mitigates reward hacking in practice across four realistic settings, including RLHF. Our code is available at https://github.com/cassidylaidlaw/orpo.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. MentalThink: Shaping Thoughts in Mental SVG World

    cs.AI 2026-07 conditional novelty 7.0 of 10

    MLLMs that generate and render SVG sketches as multi-turn intermediate reasoning steps reach 55.1% on VSIBench and 76.0% on MindCube, far above the Qwen2.5-VL-7B backbone.

  2. Do Prompt-Elicited Trajectories Reflect Training-Time Reward Hacking? A Systematic Study on Monitoring Training-Time Reward Hacking in Code Generation

    cs.LG 2026-04 unverdicted novelty 7.0 of 10

    Synthetic reward hacking data does not capture natural hacking behaviors in code generation RL, causing monitors trained on it to generalize poorly compared to those trained on in-the-wild trajectories.

  3. Bradley-Terry and Multi-Objective Reward Modeling Are Complementary

    cs.LG 2025-07 conditional novelty 6.0 of 10

    Jointly training a Bradley-Terry preference head and a multi-attribute regression head on a shared embedding improves reward-model robustness to reward hacking and boosts multi-objective scoring performance.

  4. CHiP: Cross-modal Hierarchical Direct Preference Optimization for Multimodal LLMs

    cs.CL 2025-01 conditional novelty 6.0 of 10

    CHiP adds image-level and phrase/token-level preference signals to DPO for multimodal LLMs, and the authors report large hallucination-rate reductions on Object HalBench, AMBER, MMHal, and HallusionBench.

  5. Pedagogical Safety in Educational Reinforcement Learning: Formalizing and Detecting Reward Hacking in AI Tutoring Systems

    cs.AI 2026-04 unverdicted novelty 5.5 of 10

    In a simulated AI tutor, engagement-driven RL reward-hacks; multi-objective rewards only partially help, while prerequisite and cognitive-demand constraints cut RHSI from 0.317 to 0.102.

  6. Explore or Converge? Stage-Guided Per-Step Optimization for Diffusion Models

    cs.CV 2026-08 conditional novelty 4.0 of 10

    SGPO is a stage-aware RL fine-tuning method for diffusion models that assigns a different optimization objective to each denoising stage, reducing reward hacking and improving quality, diversity, and convergence speed.

  7. Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges

    cs.AI 2025-07 reject novelty 1.0 of 10

    A broad survey of LLM alignment that catalogs objectives, benchmarks, SFT/RLHF/DPO methods, and safety challenges, without contributing new experimental or theoretical results.

Pith tools