mlsanityData to 14 Oct 2026

Why and How LLMs Hallucinate: Connecting the Dots with Subsequence Associations

arXiv 2504.12691 ↗First posted 17 Apr 2025cs.CLcited by 15

Not on the map. Why?

Abstract (arXiv)

Large language models (LLMs) frequently generate hallucinations-content that deviates from factual accuracy or provided context-posing challenges for diagnosis due to the complex interplay of underlying causes. This paper introduces a subsequence association framework to systematically trace and understand hallucinations. Our key insight is that hallucinations arise when dominant hallucinatory associations outweigh faithful ones. Through theoretical and empirical analyses, we demonstrate that decoder-only transformers effectively function as subsequence embedding models, with linear layers encoding input-output associations. We propose a tracing algorithm that identifies causal subsequences by analyzing hallucination probabilities across randomized input contexts. Experiments show our method outperforms standard attribution techniques in identifying hallucination causes and aligns with evidence from the model's training corpus. This work provides a unified perspective on hallucinations and a robust framework for their tracing and analysis.

Key facts. Cited by 15 papers in the map (data through 14 Oct 2026). Builds on 70 of its 79 references that are in the map. First posted on arXiv 17 Apr 2025.

Builds on

70 of 79 references.

References · 70

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  3. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  4. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
  5. 2014Neural Machine Translation by Jointly Learning to Align and Translatecited by 4,548
  6. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  7. 2023WebArena: A Realistic Web Environment for Building Autonomous Agentscited by 2,006
  8. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
  9. 2018Sanity Checks for Saliency Mapscited by 1,156
  10. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 1,006
  11. 2015Scheduled Sampling for Sequence Prediction with Recurrent Neural Networkscited by 803
  12. 2023The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Onlycited by 763
38 more
  1. 2022Transformers learn in-context by gradient descentcited by 732
  2. 2023Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Modelscited by 698
  3. 2024OLMo: Accelerating the Science of Language Modelscited by 613
  4. 2019ERASER: A Benchmark to Evaluate Rationalized NLP Modelscited by 530
  5. 2024Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Researchcited by 483
  6. 2023The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"cited by 407
  7. 2023Trained Transformers Learn Linear Models In-Contextcited by 337
  8. 2023Android in the Wild: A Large-Scale Dataset for Android Device Controlcited by 319
  9. 2023How Language Model Hallucinations Can Snowballcited by 311
  10. 2016Rationalizing Neural Predictionscited by 281
  11. 2023Transformers learn to implement preconditioned gradient descent for in-context learningcited by 279
  12. 2023Do Large Language Models Know What They Don't Know?cited by 243
  13. 2020Explaining by Removing: A Unified Framework for Model Explanationcited by 218
  14. 2019Probing Neural Network Comprehension of Natural Language Argumentscited by 214
  15. 2022Factuality Enhanced Language Models for Open-Ended Text Generationcited by 202
  16. 2015Visualizing and Understanding Neural Models in NLPcited by 200
  17. 2023Simple synthetic data reduces sycophancy in large language modelscited by 187
  18. 2023Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solvecited by 178
  19. 2023One Step of Gradient Descent is Provably the Optimal In-Context Learner with One Layer of Linear Self-Attentioncited by 170
  20. 2023Head-to-Tail: How Knowledgeable are Large Language Models (LLMs)? A.K.A. Will LLMs Replace Knowledge Graphs?cited by 168
  21. 2024The BrowserGym Ecosystem for Web Agent Researchcited by 161
  22. 2022The Disagreement Problem in Explainable Machine Learning: A Practitioner's Perspectivecited by 159
  23. 2021Mind the Style of Text! Adversarial and Backdoor Attacks Based on Text Style Transfercited by 158
  24. 2024OmniACT: A Dataset and Benchmark for Enabling Multimodal Generalist Autonomous Agents for Desktop and Webcited by 157
  25. 2023Investigating the Factual Knowledge Boundary of Large Language Models with Retrieval Augmentationcited by 147
  26. 2022On the Origin of Hallucinations in Conversational Models: Is it the Datasets or the Models?cited by 138
  27. 2020An Empirical Study on Robustness to Spurious Correlations using Pre-trained Language Modelscited by 134
  28. 2024How Transformers Learn Causal Structure with Gradient Descentcited by 118
  29. 2023Calibrated Language Models Must Hallucinatecited by 115
  30. 2020Learning to Faithfully Rationalize by Constructioncited by 113
  31. 2024ScreenAgent: A Vision Language Model-driven Computer Control Agentcited by 102
  32. 2019When Explanations Lie: Why Many Modified BP Attributions Failcited by 92
  33. 2023How Do Transformers Learn Topic Structure: Towards a Mechanistic Understandingcited by 86
  34. 2020Look at the First Sentence: Position Bias in Question Answeringcited by 84
  35. 2021Towards Interpreting and Mitigating Shortcut Learning Behavior of NLU Modelscited by 78
  36. 2022Impossibility Theorems for Feature Attributioncited by 78
  37. 2023Inseq: An Interpretability Toolkit for Sequence Generation Modelscited by 78
  38. 2024LLMs Will Always Hallucinate, and We Need to Live With Thiscited by 78

50 most cited of 70 All 70 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 7; 2026: 8.

15 papers.

Sort:

By area:

  1. 2025Why Language Models Hallucinatecited by 220
  2. 2025CounselBench: A Large-Scale Expert Evaluation and Adversarial Benchmarking of Large Language Models in Mental Health Question Answeringcited by 21
  3. 2025Invisible Tokens, Visible Bills: The Urgent Need to Audit Hidden Operations in Opaque LLM Servicescited by 10
  4. 2025Honesty over Accuracy: Trustworthy Language Models through Reinforced Hesitationcited by 5
  5. 2025When Bias Pretends to Be Truth: How Spurious Correlations Undermine Hallucination Detection in LLMscited by 3
  6. 2025Explainability of Large Language Models: Opportunities and Challenges toward Generating Trustworthy Explanationscited by 2
  7. 2026Hallucination Basins: A Dynamic Framework for Understanding and Controlling LLM Hallucinationscited by 2
  8. 2026The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignmentcited by 2
  9. 2026Actionable Hallucination Detection: Translating Latent Uncertainty into Agentic Critiquecited by 2
  10. 2026The Truth Stays in the Family: Enhancing Contextual Grounding via Inherited Truthful Heads in Model Lineagescited by 1
  11. 2026Understanding Why Language Models Hallucinate: Testing Reasoning Against Priorscited by 1
  12. 2025When Guessing is Rewarded: Rethinking Language Model Evaluation with Distributional Uncertainty Scoringcited by 0
3 more
  1. 2026SoftMatcha 2: A Fast and Soft Pattern Matcher for Trillion-Scale Corporacited by 0
  2. 2026When the Wrong Key Wins: Understanding and Detecting Hallucinations in LLMscited by 0
  3. 2026Tracing and Taming Hallucinated Terms via Subsequence-Association Filtering and Breadth-Aware Selection for LLM Query Expansioncited by 0