mlsanityData to 14 Oct 2026

Pride and Prejudice: LLM Amplifies Self-Bias in Self-Refinement

arXiv 2402.11436 ↗First posted 18 Feb 2024cs.CL cs.AIcited by 114

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Recent studies show that large language models (LLMs) improve their performance through self-feedback on certain tasks while degrade on others. We discovered that such a contrary is due to LLM's bias in evaluating their own output. In this paper, we formally define LLM's self-bias - the tendency to favor its own generation - using two statistics. We analyze six LLMs (GPT-4, GPT-3.5, Gemini, LLaMA2, Mixtral and DeepSeek) on translation, constrained text generation, and mathematical reasoning tasks. We find that self-bias is prevalent in all examined LLMs across multiple languages and tasks. Our analysis reveals that while the self-refine pipeline improves the fluency and understandability of model outputs, it further amplifies self-bias. To mitigate such biases, we discover that larger model size and external feedback with accurate assessment can significantly reduce bias in the self-refine pipeline, leading to actual performance improvement in downstream tasks. The code and data are released at https://github.com/xu1998hz/llm_self_bias.

Key facts. Cited by 114 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 18 of its 42 references that are in the map. First posted on arXiv 18 Feb 2024.

Builds on

18 of 42 references.

Closest ancestors · 3

Same direction, newest first

  1. 2023LLMs cannot find reasoning errors, but can correct them given the error location14 Nov
  2. 2023Large Language Models Cannot Self-Correct Reasoning Yet3 Oct
  3. 2023Automatically Correcting Large Language Models: Surveying the landscape of diverse self-correction strategies6 Aug

Other references · 15

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  3. 2024Mixtral of Expertscited by 1,823
  4. 2022No Language Left Behind: Scaling Human-Centered Machine Translationcited by 1,312
  5. 2023Gemini: A Family of Highly Capable Multimodal Modelscited by 872
  6. 2024DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Modelscited by 867
  7. 2024Self-Rewarding Language Modelscited by 681
  8. 2023Reinforced Self-Training (ReST) for Language Modelingcited by 500
  9. 2023GPTScore: Evaluate as You Desirecited by 361
  10. 2023Benchmarking Cognitive Biases in Large Language Models as Evaluatorscited by 150
  11. 2023Can Large Language Models Really Improve by Self-critiquing Their Own Plans?cited by 108
  12. 2023LLMs as Narcissistic Evaluators: When Ego Inflates Evaluation Scorescited by 90
3 more
  1. 2023GPT-4 Doesn't Know It's Wrong: An Analysis of Iterative Prompting for Reasoning Problemscited by 89
  2. 2023CritiqueLLM: Towards an Informative Critique Generation Model for Evaluation of Large Language Model Generationcited by 66
  3. 2023A Closer Look at the Self-Verification Abilities of Large Language Models in Logical Reasoningcited by 41

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 20; 2025: 49; 2026: 45.

114 papers.

Sort:

By area:

  1. 2024Justice or Prejudice? Quantifying Biases in LLM-as-a-Judgecited by 368
  2. 2024HuatuoGPT-o1, Towards Medical Complex Reasoning with LLMscited by 265
  3. 2024Self-Preference Bias in LLM-as-a-Judgecited by 258
  4. 2024Advancing LLM Reasoning Generalists with Preference Treescited by 192
  5. 2025Preference Leakage: A Contamination Problem in LLM-as-a-judgecited by 125
  6. 2025Reward Shaping to Mitigate Reward Hacking in RLHFcited by 88
  7. 2024VL-RewardBench: A Challenging Benchmark for Vision-Language Generative Reward Modelscited by 86
  8. 2025Grounded Chain-of-Thought for Multimodal Large Language Modelscited by 59
  9. 2025The FACTS Grounding Leaderboard: Benchmarking LLMs' Ability to Ground Responses to Long-Form Inputcited by 52
  10. 2025Which of These Best Describes Multiple Choice Evaluation with LLMs? A) Forced B) Flawed C) Fixable D) All of the Abovecited by 52
  11. 2024Internal Consistency and Self-Feedback in Large Language Models: A Surveycited by 50
  12. 2025Ground-R1: Incentivizing Grounded Visual Reasoning via Reinforcement Learningcited by 49
38 more
  1. 2025Training LLM-based Tutors to Improve Student Learning Outcomes in Dialoguescited by 43
  2. 2025Do LLM Evaluators Prefer Themselves for a Reason?cited by 38
  3. 2025Play Favorites: A Statistical Method to Measure Self-Bias in LLM-as-a-Judgecited by 36
  4. 2025Beyond the Surface: Measuring Self-Preference in LLM Judgmentscited by 31
  5. 2025How Is LLM Reasoning Distracted by Irrelevant Context? An Analysis Using a Controlled Benchmarkcited by 28
  6. 2024RuleArena: A Benchmark for Rule-Guided Reasoning with LLMs in Real-World Scenarioscited by 27
  7. 2025Safety Aware Task Planning via Large Language Models in Roboticscited by 24
  8. 2024From Calculation to Adjudication: Examining LLM judges on Mathematical Reasoning Taskscited by 22
  9. 2026Brittlebench: Quantifying LLM robustness via prompt sensitivitycited by 21
  10. 2024Learning Code Preference via Synthetic Evolutioncited by 19
  11. 2025Variation in Verification: Understanding Verification Dynamics in Large Language Modelscited by 19
  12. 2024Can Automatic Metrics Assess High-Quality Translations?cited by 16
  13. 2024Improving Model Factuality with Fine-grained Critique-based Evaluatorcited by 15
  14. 2025Rankers, Judges, and Assistants: Towards Understanding the Interplay of LLMs in Information Retrieval Evaluationcited by 15
  15. 2025Generative AI Act II: Test Time Scaling Drives Cognition Engineeringcited by 15
  16. 2024Depression Diagnosis Dialogue Simulation: Self-improving Psychiatrist with Tertiary Memorycited by 14
  17. 2024MIRAGE-Bench: Automatic Multilingual Benchmark Arena for Retrieval-Augmented Generation Systemscited by 14
  18. 2024Multi-expert Prompting Improves Reliability, Safety, and Usefulness of Large Language Modelscited by 14
  19. 2025Dialogue Injection Attack: Jailbreaking LLMs through Context Manipulationcited by 14
  20. 2026BiasScope: Towards Automated Detection of Bias in LLM-as-a-Judge Evaluationcited by 14
  21. 2025No Need for Explanations: LLMs can implicitly learn from mistakes in-contextcited by 13
  22. 2026Exploring Silicon-Based Societies: An Early Study of the Moltbook Agent Communitycited by 13
  23. 2024Typhoon 2: A Family of Open Text and Multimodal Thai Large Language Modelscited by 12
  24. 2025SQLens: An End-to-End Framework for Error Detection and Correction in Text-to-SQLcited by 12
  25. 2024Dialectical Alignment: Resolving the Tension of 3H and Security Threats of LLMscited by 11
  26. 2026Self-Preference Bias in Rubric-Based Evaluation of Large Language Modelscited by 11
  27. 2024PingPong: A Benchmark for Role-Playing Language Models with User Emulation and Multi-Model Evaluationcited by 10
  28. 2025CLIPPER: Compression enables long-context synthetic data generationcited by 10
  29. 2025What do Large Language Models Say About Animals? Investigating Risks of Animal Harm in Generated Textcited by 9
  30. 2025PromptPex: Automatic Test Generation for Language Model Promptscited by 9
  31. 2025SPARTA ALIGNMENT: Collectively Aligning Multiple Language Models through Combatcited by 9
  32. 2026Inherited Goal Drift: Contextual Pressure Can Undermine Agentic Goalscited by 9
  33. 2025Beyond Accuracy: The Role of Calibration in Self-Improving Large Language Modelscited by 7
  34. 2025An Empirical Study of Group Conformity in Multi-Agent Systemscited by 7
  35. 2025Agentic LLM Framework for Adaptive Decision Discoursecited by 6
  36. 2026Seven simple steps for log analysis in AI systemscited by 6
  37. 2024LLMs are Biased Evaluators But Not Biased for Retrieval Augmented Generationcited by 5
  38. 2024Visual Prompting with Iterative Refinement for Design Critique Generationcited by 5

50 most cited of 114 All 114 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024LLM Evaluators Recognize and Favor Their Own Generations
  2. 2023Self-Refine: Iterative Refinement with Self-Feedback
  3. 2024Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models
  4. 2025Can LLMs Correct Themselves? A Benchmark of Self-Correction in LLMs
  5. 2025Self-Correcting Large Language Models: Generation vs. Multiple Choice
  6. 2024Toward Self-Improvement of LLMs via Imagination, Searching, and Criticizing
  7. 2025Self-Evolving Critique Abilities in Large Language Models
  8. 2024Self-Reflection Makes Large Language Models Safer, Less Biased, and Ideologically Neutral