REVIEW 26 cited by
Secrets of RLHF in Large Language Models Part II: Reward Modeling
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
Secrets of RLHF in Large Language Models Part II: Reward Modeling
read the original abstract
Reinforcement Learning from Human Feedback (RLHF) has become a crucial technology for aligning language models with human values and intentions, enabling models to produce more helpful and harmless responses. Reward models are trained as proxies for human preferences to drive reinforcement learning optimization. While reward models are often considered central to achieving high performance, they face the following challenges in practical applications: (1) Incorrect and ambiguous preference pairs in the dataset may hinder the reward model from accurately capturing human intent. (2) Reward models trained on data from a specific distribution often struggle to generalize to examples outside that distribution and are not suitable for iterative RLHF training. In this report, we attempt to address these two issues. (1) From a data perspective, we propose a method to measure the strength of preferences within the data, based on a voting mechanism of multiple reward models. Experimental results confirm that data with varying preference strengths have different impacts on reward model performance. We introduce a series of novel methods to mitigate the influence of incorrect and ambiguous preferences in the dataset and fully leverage high-quality preference data. (2) From an algorithmic standpoint, we introduce contrastive learning to enhance the ability of reward models to distinguish between chosen and rejected responses, thereby improving model generalization. Furthermore, we employ meta-learning to enable the reward model to maintain the ability to differentiate subtle differences in out-of-distribution samples, and this approach can be utilized for iterative RLHF optimization.
Forward citations
Cited by 26 Pith papers
-
ORPO: Monolithic Preference Optimization without Reference Model
ORPO performs preference alignment during supervised fine-tuning via a monolithic odds ratio penalty, allowing 7B models to outperform larger state-of-the-art models on alignment benchmarks.
-
Towards Order Fairness: Mitigating LLMs Order Sensitivity through Dual Group Advantage Optimization
DGAO uses reinforcement learning to optimize LLMs for both accuracy and order stability by balancing intra-group accuracy advantages and inter-group stability advantages.
-
Reinforcement Learning via Value Gradient Flow
VGF solves behavior-regularized RL by transporting particles from a reference distribution to the value-induced optimal policy via discrete value-guided gradient flow.
-
Low-rank Optimization Trajectories Modeling for LLM RLVR Acceleration
NExt accelerates RLVR training for LLMs by nonlinearly extrapolating low-rank parameter trajectories extracted from LoRA runs.
-
Incentivizing High-Quality Human Annotations with Golden Questions
The paper derives a Θ(1/√(n log n)) hypothesis testing rate under strategic annotator behavior and shows that high-certainty, format-similar golden questions better reveal annotation quality than standard checks.
-
Metadata-Free Meta-Reweighted Direct Preference Optimization under Noisy Preference Labels
Bilevel reweighting recovers clean DPO optima under idealized noise, and PACMR-DPO learns sample weights without clean metadata using prompt-augmentation consistency plus LoRA finite differences.
-
Metadata-Free Meta-Reweighted Direct Preference Optimization under Noisy Preference Labels
Under 20–40% random preference-label flips, PACMR-DPO—a VNet-reweighted DPO with a prompt-augmentation-consistency meta-objective and central-difference LoRA meta-gradients—outperforms cDPO, IPO, rDPO, and Dr.DPO in j...
-
Understanding helpfulness and harmless tension in reward models
Mixed-objective reward models underperform single-objective ones because shared neurons support one objective while negatively affecting the other, creating alignment tension.
-
Representation-Aware Advantage Estimation: Your Reward Model Provides More Than A Scalar Output
GraphAE builds graphs from RM hidden-state similarities among sampled responses and propagates advantages to improve RLHF sample efficiency.
-
Boosting Self-Consistency with Ranking
RISC reformulates self-consistency answer selection as a ranking task solved by a lightweight LambdaRank model with five hand-designed features, yielding better accuracy-efficiency trade-offs than majority voting on Q...
-
Gradient-Gated DPO: Stabilizing Preference Optimization in Language Models
Gate-DPO attenuates gradients on low-probability rejected responses to reduce probability collapse and improve chosen-response likelihood during preference optimization.
-
Adaptive Margin RLHF via Preference over Preferences
Adaptive margins for DPO inferred from preference-over-preference comparisons improve alignment quality, with random sampling of comparisons working best overall.
-
How Humans Help LLMs: Assessing and Incentivizing Human Preference Annotators
Develops self-consistency monitoring for preference annotators and derives sample-complexity bounds showing linear contracts achieve near-ideal performance faster than binary ones under continuous actions.
-
Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization
Mixed Preference Optimization with the MMPR dataset boosts multimodal CoT reasoning, lifting InternVL2-8B to 67.0 accuracy on MathVista (+8.7 points) and matching the 76B model.
-
Meta-Learned Reward Shaping for Reinforcement Learning from Human Feedback
Training a small multi-task reward-shaping network and adding it to the RLHF reward is claimed to improve LLaMA-3-8B alignment across four benchmarks, but the supporting theory is not established.
-
Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning
Survey mapping RL techniques onto LLM training and highlighting gaps in value-based, off-policy, and bootstrapping methods.
-
DynaCF: Mitigating Shortcut Learning in Reward Models via Dynamic Counterfactual Sensitivity
DynaCF dynamically downweights shortcut-sensitive samples in reward model training by tracking margin shifts under online counterfactual perturbations within the Bradley-Terry loss.
-
Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance
FEST improves RLVR sample efficiency on math and coding benchmarks by combining supervised signals, on-policy signals, and decaying weights on just 128 randomly chosen demonstrations, matching full-dataset baselines.
-
Personalized Alignment Revisited: The Necessity and Sufficiency of User Diversity
A user-diversity condition is necessary and sufficient for personalized alignment to achieve O(1) online regret and log(1/epsilon) offline sample complexity.
-
DT2IT-MRM: Debiased Preference Construction and Iterative Training for Multimodal Reward Modeling
DT2IT-MRM proposes a debiased preference construction pipeline, T2I data reformulation, and iterative training to curate multimodal preference data, achieving SOTA on VL-RewardBench, Multimodal RewardBench, and MM-RLH...
-
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 op...
-
Users as Annotators: LLM Preference Learning from Comparison Mode
Introduces a latent user quality model and EM algorithm to infer and filter noisy user-provided pairwise preferences for improved LLM alignment.
-
BV-Blend: Uncertainty-Weighted Historical Baselines for Stable Critic-Free RL with Verifiable Rewards
BV-Blend blends prompt-local and semantic-cluster historical reward statistics via SEM-derived weights to stabilize critic-free RL advantage estimation.
-
Proxy Reward Internalization and Mechanistic Exploitation: A Learned Precursor to Reward Hacking and Its Generalization
Proxy RL produces a staged proxy-internalization capability that emerges before and predicts reward hacking in coding environments.
-
Seed1.5-VL Technical Report
Seed1.5-VL is a compact multimodal model that sets new records on dozens of vision-language benchmarks and outperforms prior systems on agent-style tasks.
-
Qwen2.5 Technical Report
Qwen2.5 LLMs scale pre-training data to 18 trillion tokens and apply multistage reinforcement learning, achieving competitive performance on benchmarks with models up to 5 times larger.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.