mlsanityData to 14 Oct 2026

Hallucinate at the Last in Long Response Generation: A Case Study on Long Document Summarization

arXiv 2505.15291 ↗First posted 21 May 2025cs.CLcited by 6

Not on the map. Why?

Abstract (arXiv)

Large Language Models (LLMs) have significantly advanced text generation capabilities, including tasks like summarization, often producing coherent and fluent outputs. However, faithfulness to source material remains a significant challenge due to the generation of hallucinations. While extensive research focuses on detecting and reducing these inaccuracies, less attention has been paid to the positional distribution of hallucination within generated text, particularly in long outputs. In this work, we investigate where hallucinations occur in LLM-based long response generation, using long document summarization as a key case study. Focusing on the challenging setting of long context-aware long response generation, we find a consistent and concerning phenomenon: hallucinations tend to concentrate disproportionately in the latter parts of the generated long response. To understand this bias, we explore potential contributing factors related to the dynamics of attention and decoding over long sequences. Furthermore, we investigate methods to mitigate this positional hallucination, aiming to improve faithfulness specifically in the concluding segments of long outputs.

Key facts. Cited by 6 papers in the map (data through 14 Oct 2026). Builds on 45 of its 59 references that are in the map. First posted on arXiv 21 May 2025.

Builds on

45 of 59 references.

References · 45

  1. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  2. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  3. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  4. 2023A Survey on Large Language Model based Autonomous Agentscited by 2,242
  5. 2025Gemma 3 Technical Reportcited by 1,824
  6. 2019Generating Long Sequences with Sparse Transformerscited by 1,546
  7. 2018Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarizationcited by 1,317
  8. 2023FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generationcited by 1,282
  9. 2024ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Toolscited by 1,121
  10. 2015Teaching Machines to Read and Comprehendcited by 1,049
  11. 2018A Discourse-Aware Attention Model for Abstractive Summarization of Long Documentscited by 478
  12. 2023Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuningcited by 445
33 more
  1. 2024MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attentioncited by 383
  2. 2021Efficient Attentions for Long Document Summarizationcited by 345
  3. 2023Trusting Your Evidence: Hallucinate Less with Context-aware Decodingcited by 327
  4. 2024MiniCheck: Efficient Fact-Checking of LLMs on Grounding Documentscited by 257
  5. 2024LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMscited by 144
  6. 2024Lookback Lens: Detecting and Mitigating Contextual Hallucinations in Large Language Models Using Only Attention Mapscited by 142
  7. 2023BooookScore: A systematic exploration of book-length summarization in the era of LLMscited by 137
  8. 2024RLAIF-V: Open-Source AI Feedback Leads to Super GPT-4V Trustworthinesscited by 131
  9. 2024A Thorough Examination of Decoding Methods in the Era of LLMscited by 114
  10. 2022Understanding Factual Errors in Summarization: Errors, Summarizers, Datasets, Error Detectorscited by 104
  11. 2018Abstractive Summarization of Reddit Posts with Multi-level Memory Networkscited by 102
  12. 2024Found in the Middle: Calibrating Positional Attention Bias Improves Long Context Utilizationcited by 101
  13. 2023Tell Your Model Where to Attend: Post-hoc Attention Steering for LLMscited by 91
  14. 2024FineSurE: Fine-grained Summarization Evaluation using LLMscited by 83
  15. 2023Factually Consistent Summarization via Reinforcement Learning with Textual Entailment Feedbackcited by 72
  16. 2024FABLES: Evaluating faithfulness and content selection in book-length summarizationcited by 65
  17. 2024LongGenBench: Benchmarking Long-Form Generation in Long Context LLMscited by 47
  18. 2024AdaCAD: Adaptively Decoding to Balance Conflicts between Contextual and Parametric Knowledgecited by 37
  19. 2023Merging Generated and Retrieved Knowledge for Open-Domain QAcited by 36
  20. 2024FENICE: Factuality Evaluation of summarization based on Natural language Inference and Claim Extractioncited by 28
  21. 2024Learning to Summarize from LLM-generated Feedbackcited by 24
  22. 2024Fine-Grained Natural Language Inference Based Faithfulness Evaluation for Diverse Summarisation Taskscited by 23
  23. 2024Mitigating Hallucinations in Large Vision-Language Models via Summary-Guided Decodingcited by 22
  24. 2024On Positional Bias of Faithfulness for Long-form Summarizationcited by 22
  25. 2024Toward Robust Hyper-Detailed Image Captioning: A Multiagent Approach and Dual Evaluation Metrics for Factuality and Coveragecited by 20
  26. 2024From Single to Multi: How LLMs Hallucinate in Multi-Document Summarizationcited by 17
  27. 2023LongDocFACTScore: Evaluating the Factuality of Long Document Abstractive Summarisationcited by 16
  28. 2024FIZZ: Factual Inconsistency Detection by Zoom-in Summary and Zoom-out Documentcited by 15
  29. 2025Shifting Long-Context LLMs Research from Input to Outputcited by 14
  30. 2021On-the-Fly Attention Modulation for Neural Generationcited by 12
  31. 2025Discourse-Driven Evaluation: Unveiling Factual Inconsistency in Long Document Summarizationcited by 8
  32. 2023HistAlign: Improving Context Dependency in Language Generation by Aligning with Historycited by 6
  33. 2022NarraSum: A Large-Scale Dataset for Abstractive Narrative Summarizationcited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 1; 2026: 5.

6 papers.

  1. 2025The Curious Case of Factual (Mis)Alignment between LLMs' Short- and Long-Form Answerscited by 2
  2. 2026From 'May' to 'Is': Certainty Distortion in Language Model Rewritingcited by 0
  3. 2026On-Device Deep Research at 4B: Exposure Bounds Faithfulness, Retrieval Bounds Coveragecited by 0
  4. 2026Reliability Scales Inversely: Hallucinations Snowball Faster in Bigger Language Modelscited by 0
  5. 2026Doc2FRC: Length-Consistent Document-Level Machine Translation via Fixed-Range Chunkingcited by 0
  6. 2026From Utterances to Networks: Modelling Slang Adoption and Diffusion Across Subredditscited by 0