mlsanityData to 7 Oct 2026

Detoxifying Large Language Models via Knowledge Editing

arXiv 2403.14472 ↗First posted 21 Mar 2024cs.CL cs.AI cs.CVcited by 86

Not on the map. Why?

Cited by 26 in LLM safety and security, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

This paper investigates using knowledge editing techniques to detoxify Large Language Models (LLMs). We construct a benchmark, SafeEdit, which covers nine unsafe categories with various powerful attack prompts and equips comprehensive metrics for systematic evaluation. We conduct experiments with several knowledge editing approaches, indicating that knowledge editing has the potential to detoxify LLMs with a limited impact on general performance efficiently. Then, we propose a simple yet effective baseline, dubbed Detoxifying with Intraoperative Neural Monitoring (DINM), to diminish the toxicity of LLMs within a few tuning steps via only one instance. We further provide an in-depth analysis of the internal mechanism for various detoxifying approaches, demonstrating that previous methods like SFT and DPO may merely suppress the activations of toxic parameters, while DINM mitigates the toxicity of the toxic parameters to a certain extent, making permanent adjustments. We hope that these insights could shed light on future work of developing detoxifying approaches and the underlying knowledge mechanisms of LLMs. Code and benchmark are available at https://github.com/zjunlp/EasyEdit.

Key facts. Cited by 86 papers in the map (data through 7 Oct 2026). Builds on 64 of its 93 references that are in the map. First posted on arXiv 21 Mar 2024.

Builds on

64 of 93 references.

References · 64

  1. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  2. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  3. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  4. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,895
  5. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  6. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,565
  7. 2018Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarizationcited by 1,308
  8. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,239
  9. 2022Mass-Editing Memory in a Transformercited by 976
  10. 2022Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Spacecited by 522
  11. 2023GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Promptscited by 520
  12. 2023DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Modelscited by 505
38 more
  1. 2023Jailbreaking ChatGPT via Prompt Engineering: An Empirical Studycited by 478
  2. 2022Memory-Based Model Editing at Scalecited by 443
  3. 2023Catastrophic Jailbreak of Open-source LLMs via Exploiting Generationcited by 439
  4. 2023SmoothLLM: Defending Large Language Models Against Jailbreaking Attackscited by 434
  5. 2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrationscited by 401
  6. 2023Editing Large Language Models: Problems, Methods, and Opportunitiescited by 367
  7. 2023Toxicity in ChatGPT: Analyzing Persona-assigned Language Modelscited by 356
  8. 2023MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questionscited by 292
  9. 2023Can We Edit Factual Knowledge by In-Context Learning?cited by 284
  10. 2022Aging with GRACE: Lifelong Model Editing with Discrete Key-Value Adaptorscited by 267
  11. 2023LEACE: Perfect linear concept erasure in closed formcited by 254
  12. 2023A Wolf in Sheep's Clothing: Generalized Nested Jailbreak Prompts can Fool Large Language Models Easilycited by 242
  13. 2023Evaluating the Ripple Effects of Knowledge Editing in Language Modelscited by 238
  14. 2023Does Localization Inform Editing? Surprising Differences in Causality-Based Localization vs. Knowledge Editing in Language Modelscited by 237
  15. 2023Transformer-Patcher: One Mistake worth One Neuroncited by 196
  16. 2023Knowledge Editing for Large Language Models: A Surveycited by 189
  17. 2024A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicitycited by 179
  18. 2023Defending Against Alignment-Breaking Attacks via Robustly Aligned LLMcited by 173
  19. 2023PMET: Precise Model Editing in a Transformercited by 170
  20. 2024Secrets of RLHF in Large Language Models Part II: Reward Modelingcited by 170
  21. 2023Defending Large Language Models Against Jailbreaking Attacks Through Goal Prioritizationcited by 156
  22. 2024Model Editing at Scale leads to Gradual and Catastrophic Forgettingcited by 92
  23. 2023MELO: Enhancing Model Editing with Neuron-Indexed Dynamic LoRAcited by 80
  24. 2020Back to the Future: Unsupervised Backprop-based Decoding for Counterfactual and Abductive Commonsense Reasoningcited by 65
  25. 2023Journey to the Center of the Knowledge Neurons: Discoveries of Language-Independent Knowledge Neurons and Degenerate Knowledge Neuronscited by 56
  26. 2023From Instructions to Intrinsic Human Values -- A Survey of Alignment Goals for Big Modelscited by 45
  27. 2023Self-Detoxifying Language Models via Toxification Reversalcited by 45
  28. 2023Testing Language Model Agents Safely in the Wildcited by 45
  29. 2023Using In-Context Learning to Improve Dialogue Safetycited by 41
  30. 2022Probing Classifiers are Unreliable for Concept Removal and Detectioncited by 40
  31. 2022Detoxifying Text with MaRCo: Controllable Revision with Experts and Anti-Expertscited by 36
  32. 2023Finding and Editing Multi-Modal Neurons in Pre-Trained Transformerscited by 36
  33. 2023A Survey on Knowledge Editing of Neural Networkscited by 34
  34. 2023Separate the Wheat from the Chaff: Model Deficiency Unlearning via Parameter-Efficient Module Operationcited by 33
  35. 2024DeepEdit: Knowledge Editing as Decoding with Constraintscited by 33
  36. 2023Trends in Integration of Knowledge and Large Language Models: A Survey and Taxonomy of Methods, Benchmarks, and Applicationscited by 28
  37. 2023Emptying the Ocean with a Spoon: Should We Edit Models?cited by 27
  38. 2022Language Anisotropic Cross-Lingual Model Editingcited by 26

50 most cited of 64 All 64 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 34; 2025: 37; 2026: 15.

86 papers.

Sort:

By area:

  1. 2024A Survey on the Memory Mechanism of Large Language Model based Agentscited by 622
  2. 2025A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deploymentcited by 110
  3. 2024Optimus-1: Hybrid Multimodal Memory Empowered Agents Excel in Long-Horizon Taskscited by 83
  4. 2024FlipAttack: Jailbreak LLMs via Flippingcited by 73
  5. 2024When LLMs Meet Cybersecurity: A Systematic Literature Reviewcited by 70
  6. 2024JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Modelscited by 67
  7. 2024Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Trainingcited by 67
  8. 2024Navigating the Risks: A Survey of Security, Privacy, and Ethics Threats in LLM-Based Agentscited by 63
  9. 2024Caution for the Environment: Multimodal LLM Agents are Susceptible to Environmental Distractionscited by 55
  10. 2024Knowledge Mechanisms in Large Language Models: A Survey and Perspectivecited by 50
  11. 2024Model Editing as a Robust and Denoised variant of DPO: A Case Study on Toxicitycited by 33
  12. 2024In-Context Editing: Learning Knowledge from Self-Induced Distributionscited by 33
38 more
  1. 2024Preference Tuning For Toxicity Mitigation Generalizes Across Languagescited by 31
  2. 2024Can Knowledge Editing Really Correct Hallucinations?cited by 28
  3. 2024To Forget or Not? Towards Practical Knowledge Unlearning for Large Language Modelscited by 25
  4. 2024Beyond Single Concept Vector: Modeling Concept Subspace in LLMs with Gaussian Distributioncited by 24
  5. 2025Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atomscited by 23
  6. 2024Can Editing LLMs Inject Harm?cited by 22
  7. 2024Locking Down the Finetuned LLMs Safetycited by 22
  8. 2025Reasoning as an Adaptive Defense for Safetycited by 22
  9. 2024PaCE: Parsimonious Concept Engineering for Large Language Modelscited by 19
  10. 2024Unbridled Icarus: A Survey of the Potential Perils of Image Inputs in Multimodal Large Language Model Securitycited by 18
  11. 2024Mitigating the Language Mismatch and Repetition Issues in LLM-based Machine Translation via Model Editingcited by 18
  12. 2024From RAGs to rich parameters: Probing how language models utilize external knowledge over parametric information for factual queriescited by 16
  13. 2024Unveiling Entity-Level Unlearning for Large Language Models: A Comprehensive Analysiscited by 14
  14. 2024Lifelong Knowledge Editing for Vision Language Models with Low-Rank Mixture-of-Expertscited by 14
  15. 2025Interpretation Meets Safety: A Survey on Interpretation Methods and Tools for Improving LLM Safetycited by 10
  16. 2024Decoupled Alignment for Robust Plug-and-Play Adaptationcited by 9
  17. 2025Detoxifying Large Language Models via Autoregressive Reward Guided Representation Editingcited by 9
  18. 2024O-Edit: Orthogonal Subspace Editing for Language Model Sequential Editingcited by 8
  19. 2025Bring Your Own Knowledge: A Survey of Methods for LLM Knowledge Expansioncited by 8
  20. 2025Implicit Bias in LLMs: A Surveycited by 8
  21. 2025UniDetox: Universal Detoxification of Large Language Models via Dataset Distillationcited by 8
  22. 2025Knowledge Fusion of Large Language Models Via Modular SkillPackscited by 8
  23. 2024Identifying Knowledge Editing Types in Large Language Modelscited by 7
  24. 2025Activation Approximations Can Incur Safety Vulnerabilities Even in Aligned LLMs: Comprehensive Analysis and Defensecited by 7
  25. 2025LED-Merging: Mitigating Safety-Utility Conflicts in Model Merging with Location-Election-Disjointcited by 7
  26. 2025EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Modelscited by 7
  27. 2025UltraEdit: Training-, Subject-, and Memory-Free Lifelong Editing in Language Modelscited by 7
  28. 2025IF-GUIDE: Influence Function-Guided Detoxification of LLMscited by 7
  29. 2025Layer-Level Self-Exposure and Patch: Affirmative Token Mitigation for Jailbreak Attack Defensecited by 6
  30. 2025GeoEdit: Geometric Knowledge Editing for Large Language Modelscited by 6
  31. 2026From Data to Behavior: Predicting Unintended Model Behaviors Before Trainingcited by 6
  32. 2024Can LLMs be Fooled? Investigating Vulnerabilities in LLMscited by 5
  33. 2025Aligned Probing: Relating Toxic Behavior and Model Internalscited by 5
  34. 2025Revealing the Deceptiveness of Knowledge Editing: A Mechanistic Analysis of Superficial Editingcited by 5
  35. 2025Adaptive Detoxification: Safeguarding General Capabilities of LLMs through Toxicity-Aware Knowledge Editingcited by 5
  36. 2025Large Language Model Sourcing: A Surveycited by 5
  37. 2025SAKE: Towards Editing Auditory Attribute Knowledge of Large Audio-Language Modelscited by 5
  38. 2024ContraSolver: Self-Alignment of Language Models by Resolving Internal Preference Contradictionscited by 4

50 most cited of 86 All 86 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Model Surgery: Modulating LLM's Behavior Via Simple Parameter Editing
  2. 2025SGM: Safety Glasses for Multimodal Large Language Models via Neuron-Level Detoxification
  3. 2024Model Editing Harms General Abilities of Large Language Models: Regularization to the Rescue
  4. 2024Large Language Models can be Strong Self-Detoxifiers
  5. 2026Detoxifying LLMs via Representation Erasure-Based Preference Optimization
  6. 2026Where Does Toxicity Live? Mechanistic Localization and Targeted Suppression in Language Models
  7. 2024Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning
  8. 2025<think> So let's replace this phrase with insult... </think> Lessons learned from generation of toxic texts with LLMs