Pith. sign in

REVIEW 17 cited by

RRHF: Rank Responses to Align Language Models with Human Feedback without tears

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 2304.05302 v3 pith:S4YJM6DB submitted 2023-04-11 cs.CL

classification cs.CL
keywords rrhfhumanmodelmodelsresponseslanguagealignpreferences
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Reinforcement Learning from Human Feedback (RLHF) facilitates the alignment of large language models with human preferences, significantly enhancing the quality of interactions between humans and models. InstructGPT implements RLHF through several stages, including Supervised Fine-Tuning (SFT), reward model training, and Proximal Policy Optimization (PPO). However, PPO is sensitive to hyperparameters and requires multiple models in its standard implementation, making it hard to train and scale up to larger parameter counts. In contrast, we propose a novel learning paradigm called RRHF, which scores sampled responses from different sources via a logarithm of conditional probabilities and learns to align these probabilities with human preferences through ranking loss. RRHF can leverage sampled responses from various sources including the model responses from itself, other large language model responses, and human expert responses to learn to rank them. RRHF only needs 1 to 2 models during tuning and can efficiently align language models with human preferences robustly without complex hyperparameter tuning. Additionally, RRHF can be considered an extension of SFT and reward model training while being simpler than PPO in terms of coding, model counts, and hyperparameters. We evaluate RRHF on the Helpful and Harmless dataset, demonstrating comparable alignment performance with PPO by reward model score and human labeling. Extensive experiments show that the performance of RRHF is highly related to sampling quality which suggests RRHF is a best-of-n learner. Codes available at https://github.com/GanjinZero/RRHF.

Discussion (0). Sign in to comment.

Forward citations

Cited by 17 Pith papers

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

  1. Test-Time Scaling via Error Localization

    cs.LG 2026-07 conditional novelty 6.0 of 10

    TTEL uses feedback-induced token probability drops to localize the first error in a failed reasoning trace and branch a new generation from that prefix, improving pass@k per token on coding and math benchmarks.

  2. Every Sample Counts: Supervised Fine-Tuning of Language Models with Pointwise Constraints

    eess.SP 2026-07 conditional novelty 6.0 of 10

    Pointwise constrained fine-tuning via sample-wise augmented Lagrangians and learned relaxations reduces tail constraint violations across safety, tool-calling, and re-ranking while preserving average task performance.

  3. Multi-Turn On-Policy Distillation with Prefix Replay

    cs.LG 2026-07 conditional novelty 6.0 of 10

    ReOPD offline-distills multi-turn agentic LLMs via teacher-prefix replay plus step-decay sampling, matching online OPD accuracy at ≥4× speed with zero tool calls.

  4. SCOPE and SCION: A Benchmark and an Auditable Reference Pipeline for Schema Induction and Fusion from Text

    cs.AI 2026-05 conditional novelty 6.0 of 10

    A 24-dataset benchmark for inducing schema graphs from raw text, plus an auditable LLM-based pipeline that reports the highest scores on the benchmark's four schema-similarity metrics.

  5. RLearner-LLM: Balancing Logical Grounding and Fluency in Large Language Models via Hybrid Direct Preference Optimization

    cs.CL 2026-05 unverdicted novelty 6.0 of 10

    RLearner-LLM's Hybrid-DPO fuses DeBERTa NLI and LLM verifier scores to deliver up to 6x higher NLI entailment than standard SFT while preserving answer coverage across academic domains.

  6. Simultaneous Multi-objective Alignment Across Verifiable and Non-verifiable Rewards

    cs.LG 2025-10 conditional novelty 6.0 of 10

    MAHALO aligns LLMs to multiple objectives in one model via per-objective action heads and PRM-guided decoding, improving math, value, and tutoring metrics jointly.

  7. Secure Tug-of-War (SecTOW): Iterative Defense-Attack Training with Reinforcement Learning for Multimodal Model Security

    cs.CR 2025-07 conditional novelty 6.0 of 10

    An iterative attacker-defender reinforcement learning method that makes a multimodal LLM refuse more jailbreak prompts without over-refusing ordinary queries.

  8. Enhancing Safe and Controllable Protein Generation via Knowledge Preference Optimization

    cs.AI 2025-07 conditional novelty 6.0 of 10

    A knowledge-graph-guided preference optimization framework that fine-tunes protein language models to generate fewer sequences similar to known harmful proteins.

  9. Flipping Knowledge Distillation: Leveraging Small Models' Expertise to Enhance LLMs in Text Matching

    cs.CL 2025-07 conditional novelty 6.0 of 10

    A flipped distillation method lets a decoder-only LLM learn text-matching similarity from a smaller encoder teacher through LoRA and a margin-aware contrastive loss, improving matching accuracy and online FAQ retrieval.

  10. From General to Targeted Rewards: Surpassing GPT-4 in Open-Ended Long-Context Generation

    cs.CL 2025-06 conditional novelty 6.0 of 10

    ProxyReward trains long-form generation models by rewarding how well an AI judge can answer generated yes/no questions about the response, improving open-source models on ProxyQA.

  11. Quo Vadis, World Modeling?

    cs.CV 2026-08 conditional novelty 5.0 of 10

    An agent-centric reframing of world modeling, replacing physical state prediction with 'information transitions' organized into six proxy functions and three empowerment levels.

  12. Beyond Surface-Level Detection: Towards Cognitive-Driven Defense Against Jailbreak Attacks via Meta-Operations Reasoning

    cs.AI 2025-08 unverdicted novelty 5.0 of 10

    A jailbreak defense that reasons about hidden manipulations in attack prompts, trained with supervised fine-tuning plus entropy-guided reinforcement learning, generalizes to attacks never seen in training.

  13. Activation Reward Models for Few-Shot Model Alignment

    cs.CV 2025-07 conditional novelty 5.0 of 10

    Mean attention-head activations from a few labeled examples, injected into selected heads, turn a frozen vision-language model into a few-shot reward model that beats prompting and scoring baselines and a new reward-h...

  14. Value-Free Policy Optimization via Reward Partitioning

    cs.LG 2025-06 conditional novelty 5.0 of 10

    RPO regresses the policy log-ratio against rewards centered by an empirical partition function estimated from same-prompt responses, removing the value network and reporting improved alignment, diversity, and lower toxicity.

  15. PKG-DPO: Optimizing Domain-Specific AI systems with Physics Knowledge Graphs and Direct Preference Optimization

    cs.AI 2025-08 reject novelty 3.0 of 10

    A physics knowledge graph plus Direct Preference Optimization reduces physically invalid AI recommendations in welding by 17% over knowledge-graph-only DPO, but the evidence is weakened by circular evaluation and no p...

  16. 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.

  17. Inverse Reinforcement Learning Meets Large Language Model Post-Training: Basics, Advances, and Opportunities

    cs.LG 2025-07 unverdicted novelty 1.0 of 10

    A tutorial reviewing LLM alignment through the lens of inverse reinforcement learning, arguing that neural reward models learned from human data are central to post-training.

Pith tools