Pith. sign in

REVIEW 12 cited by

RAIN: Your Language Models Can Align Themselves without Finetuning

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

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 2309.07124 v2 pith:DBWPQYME submitted 2023-09-13 cs.CL

Yuhui Li , Fangyun Wei , Jinjing Zhao , Chao Zhang , Hongyang Zhang This is my paper · ORCID
classification cs.CL
keywords rainllmsdatahumaninferencemodelswithoutalignment
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Large language models (LLMs) often demonstrate inconsistencies with human preferences. Previous research typically gathered human preference data and then aligned the pre-trained models using reinforcement learning or instruction tuning, a.k.a. the finetuning step. In contrast, aligning frozen LLMs without requiring alignment data is more appealing. This work explores the potential of the latter setting. We discover that by integrating self-evaluation and rewind mechanisms, unaligned LLMs can directly produce responses consistent with human preferences via self-boosting. We introduce a novel inference method, Rewindable Auto-regressive INference (RAIN), that allows pre-trained LLMs to evaluate their own generation and use the evaluation results to guide rewind and generation for AI safety. Notably, RAIN operates without the need of extra data for model alignment and abstains from any training, gradient computation, or parameter updates. Experimental results evaluated by GPT-4 and humans demonstrate the effectiveness of RAIN: on the HH dataset, RAIN improves the harmlessness rate of LLaMA 30B from 82% of vanilla inference to 97%, while maintaining the helpfulness rate. On the TruthfulQA dataset, RAIN improves the truthfulness of the already-well-aligned LLaMA-2-chat 13B model by 5%.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 12 Pith papers

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

  1. Kwai Keye-VL 1.5 Technical Report

    cs.CV 2025-09 conditional novelty 6.0 of 10

    Keye-VL-1.5 combines similarity-based Slow-Fast video token allocation with progressive context extension and iterative RL, reporting leading video-understanding results among 8B-scale multimodal models.

  2. PUZZLED: Jailbreaking LLMs through Word-Based Puzzles

    cs.AI 2025-08 unverdicted novelty 6.0 of 10

    PUZZLED masks harmful keywords as word-search, anagram, or crossword puzzles and achieves a reported 88.8% average attack success rate across five leading LLMs.

  3. DenseWorld-1M: Towards Detailed Dense Grounded Caption in the Real World

    cs.CV 2025-06 conditional novelty 6.0 of 10

    DenseWorld-1M provides one million images with detailed object captions, pixel masks, and spatial relations by chaining SAM, APE, RAM++, and VLMs through a three-stage labeling pipeline.

  4. Text-Diffusion Red-Teaming of Large Language Models: Unveiling Harmful Behaviors with Proximity Constraints

    cs.LG 2025-01 conditional novelty 6.0 of 10

    DART, a single-step embedding-space perturber trained with reinforcement learning, finds toxic prompts closer to reference prompts than fine-tuned or few-shot baselines on three LLMs.

  5. Stream Aligner: Efficient Sentence-Level Alignment via Distribution Induction

    cs.CL 2025-01 reject novelty 6.0 of 10

    A sentence-level iterative correction method is claimed to improve large language model alignment and math reasoning, but the evidence is compromised by GPT-4 serving as both trainer and judge.

  6. SafeLLM: Unlearning Harmful Outputs from Large Language Models against Jailbreak Attacks

    cs.LG 2025-08 unverdicted novelty 5.0 of 10

    SafeLLM detects unsafe outputs, traces them to specific feedforward-network components, and applies constrained optimization to unlearn harmful generation while preserving general capability.

  7. Self-Instruct Few-Shot Jailbreaking: Decompose the Attack into Pattern and Behavior Learning

    cs.AI 2025-01 conditional novelty 5.0 of 10

    A few-shot jailbreak method that combines repeated special-token patterns with self-generated harmful demos to push sample-level attack success near 90% on several open-source LLMs.

  8. Layer-Level Self-Exposure and Patch: Affirmative Token Mitigation for Jailbreak Attack Defense

    cs.CR 2025-01 reject novelty 5.0 of 10

    Layer-AdvPatcher edits 'toxic' transformer layers using self-generated harmful examples to block jailbreaks, but its reported attack-success rates worsen on several benchmarks.

  9. Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignment

    cs.CR 2024-11 conditional novelty 5.0 of 10

    An inference-time alignment method using a safety reward model and controlled decoding that reduces jailbreak success rates in multimodal LLMs while preserving utility.

  10. Refining Positive and Toxic Samples for Dual Safety Self-Alignment of LLMs with Minimal Human Interventions

    cs.CL 2025-02 conditional novelty 4.0 of 10

    PT-ALIGN uses self-generated positive and toxic sample pairs with MLE plus fine-grained token-level unlikelihood training to improve LLM safety with minimal human annotation.

  11. A Survey on Responsible LLMs: Inherent Risk, Malicious Use, and Mitigation Strategy

    cs.AI 2025-01 conditional novelty 4.0 of 10

    A survey that organizes responsible-LLM research into five risk dimensions and four intervention phases, reviewing privacy, hallucination, value, toxicity, and jailbreak mitigation.

  12. Search, Verify and Feedback: Towards Next Generation Post-training Paradigm of Foundation Models via Verifier Engineering

    cs.AI 2024-11 conditional novelty 4.0 of 10

    A survey and position paper that reframes post-training of foundation models as a search, verify, and feedback pipeline called verifier engineering.

Pith tools