mlsanityData to 13 Oct 2026

Diagnosing Knowledge Conflict in Multimodal Long-Chain Reasoning

arXiv 2602.14518 ↗First posted 16 Feb 2026cs.AIcited by 1

On the map: LLM reasoning → Reasoning benchmarks and evaluation

also in 1 more
Abstract (arXiv)

Multimodal large language models (MLLMs) in long chain-of-thought reasoning often fail when different knowledge sources provide conflicting signals. We formalize these failures under a unified notion of knowledge conflict, distinguishing input-level objective conflict from process-level effective conflict. Through probing internal representations, we reveal that: (I) Linear Separability: different conflict types are explicitly encoded as linearly separable features rather than entangled; (II) Depth Localization: conflict signals concentrate in mid-to-late layers, indicating a distinct processing stage for conflict encoding; (III) Hierarchical Consistency: aggregating noisy token-level signals along trajectories robustly recovers input-level conflict types; and (IV) Directional Asymmetry: reinforcing the model's implicit source preference under conflict is far easier than enforcing the opposite source. Our findings provide a mechanism-level view of multimodal reasoning under knowledge conflict and enable principled diagnosis and control of long-CoT failures.

Key facts. Cited by 1 paper in the map (data through 13 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 57 of its 60 references that are in the map. First posted on arXiv 16 Feb 2026.

Builds on

57 of 60 references.

References · 57

  1. 2025Qwen2.5-VL Technical Reportcited by 5,499
  2. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,949
  3. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,799
  4. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,101
  5. 2023Towards Understanding Sycophancy in Language Modelscited by 1,018
  6. 2024Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMscited by 827
  7. 2024Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMscited by 723
  8. 2023Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decodingcited by 677
  9. 2023HallusionBench: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Modelscited by 577
  10. 2024GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Modelscited by 542
  11. 2024MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmarkcited by 502
  12. 2024LLaVA-CoT: Let Vision Language Models Reason Step-by-Stepcited by 449
38 more
  1. 2023RLHF-V: Towards Trustworthy MLLMs via Behavior Alignment from Fine-grained Correctional Human Feedbackcited by 407
  2. 2025R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalizationcited by 366
  3. 2024A Survey on Hallucination in Large Vision-Language Modelscited by 352
  4. 2023Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflictscited by 337
  5. 2024Prismatic VLMs: Investigating the Design Space of Visually-Conditioned Language Modelscited by 337
  6. 2023How Language Model Hallucinations Can Snowballcited by 310
  7. 2024MM-LLMs: Recent Advances in MultiModal Large Language Modelscited by 278
  8. 2025Multimodal Chain-of-Thought Reasoning: A Comprehensive Surveycited by 178
  9. 2024The Revolution of Multimodal Large Language Models: A Surveycited by 135
  10. 2024RLAIF-V: Open-Source AI Feedback Leads to Super GPT-4V Trustworthinesscited by 129
  11. 2025A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deploymentcited by 114
  12. 2024ClashEval: Quantifying the tug-of-war between an LLM's internal prior and external evidencecited by 112
  13. 2025SycEval: Evaluating LLM Sycophancycited by 107
  14. 2024Efficient Multimodal Large Language Models: A Surveycited by 95
  15. 2025Words or Vision: Do Vision-Language Models Have Blind Faith in Text?cited by 87
  16. 2024WildVision: Evaluating Vision-Language Models in the Wild with Human Preferencescited by 76
  17. 2024ConflictBank: A Benchmark for Evaluating the Influence of Knowledge Conflicts in LLMcited by 73
  18. 2025Virgo: A Preliminary Exploration on Reproducing o1-like MLLMcited by 68
  19. 2023Massive Editing for Large Language Models via Meta Learningcited by 62
  20. 2024PhD: A ChatGPT-Prompted Visual hallucination Evaluation Datasetcited by 45
  21. 2024Monitoring Latent World States in Language Models with Propositional Probescited by 41
  22. 2024IRCAN: Mitigating Knowledge Conflicts in LLM Generation via Identifying and Reweighting Context-Aware Neuronscited by 37
  23. 2024Insight Over Sight: Exploring the Vision-Knowledge Conflicts in Multimodal LLMscited by 37
  24. 2025Real-Time Detection of Hallucinated Entities in Long-Form Generationcited by 34
  25. 2025Evaluating and Steering Modality Preferences in Multimodal Large Language Modelcited by 33
  26. 2024Unveiling the Ignorance of MLLMs: Seeing Clearly, Answering Incorrectlycited by 29
  27. 2024Task Preference Optimization: Improving Multimodal Large Language Models with Vision Task Alignmentcited by 29
  28. 2025Taming Knowledge Conflicts in Language Modelscited by 28
  29. 2025FaithfulRAG: Fact-Level Conflict Modeling for Context-Faithful Retrieval-Augmented Generationcited by 28
  30. 2024To Trust or Not to Trust? Enhancing Large Language Models' Situated Faithfulness to External Contextscited by 25
  31. 2023Editing Language Model-based Knowledge Graph Embeddingscited by 23
  32. 2023UNK-VQA: A Dataset and a Probe into the Abstention Ability of Multi-modal Large Modelscited by 23
  33. 2025Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-trainingcited by 23
  34. 2025MMKE-Bench: A Multimodal Editing Benchmark for Diverse Visual Knowledgecited by 22
  35. 2025KARL: Knowledge-Aware Reasoning and Reinforcement Learning for Knowledge-Intensive Visual Groundingcited by 21
  36. 2024KnowPO: Knowledge-aware Preference Optimization for Controllable Knowledge Selection in Retrieval-Augmented Language Modelscited by 20
  37. 2025Probe before You Talk: Towards Black-box Defense against Backdoor Unalignment for Large Language Modelscited by 18
  38. 2024Visual-O1: Understanding Ambiguous Instructions via Multi-modal Multi-turn Chain-of-thoughts Reasoningcited by 17

50 most cited of 57 All 57 (JSON)

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Senses Wide Shut: A Representation-Action Gap in Omnimodal LLMscited by 2

Similar, not cited

Close in wording, no citation either way

  1. 2025CrossCheck-Bench: Diagnosing Compositional Failures in Multimodal Conflict Resolution
  2. 2026Large Language Models in Resolving Contextual Knowledge Conflicts
  3. 2025Benchmarking Multimodal Knowledge Conflict for Large Multimodal Models
  4. 2024Untangle the KNOT: Interweaving Conflicting Knowledge and Reasoning Skills in Large Language Models
  5. 2026FlowEdit: Information-Theoretic Control of LLM Reasoning Flows for Ill-posed Problems Involving Conflicts
  6. 2025Compose and Fuse: Revisiting the Foundational Bottlenecks in Multimodal Reasoning
  7. 2026Compliance vs. Sensibility: On the Reasoning Controllability in Large Language Models
  8. 2026Exploring Knowledge Conflicts for Faithful LLM Reasoning: Benchmark and Method