pith. sign in

arxiv: 2607.02052 · v1 · pith:BN736OA2new · submitted 2026-07-02 · 💻 cs.SE

Mitigating Package Hallucinations in Large Language Models via Model Editing

Shuhan Liu , Yukai Zhao , Xing Hu , Kui Liu , Xiaohu Yang , Xin Xia This is my paper

Pith reviewed 2026-07-03 08:51 UTC · model grok-4.3

classification 💻 cs.SE
keywords package hallucinationmodel editinglarge language modelssoftware engineeringLoRA adaptersboundary editingcode generationsupply chain attacks
0
0 comments X

The pith

BOUND edits specific LLM modules with a boundary-aware objective to cut package hallucinations while keeping valid recommendations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces BOUND as a lightweight localized editing method that treats package hallucination as a problem of refining the model's internal boundary between valid and invalid package names in a given task context. It first uses risk-aware localization to find the relevant modules, then applies LoRA adapters trained with an objective that strengthens correct packages, weakens hallucinated ones, and maintains behavior on unrelated inputs. Experiments demonstrate large drops in hallucination rates for package recommendation on both the prompts used for editing and entirely new prompts, with the same boundary carrying over to lower errors in code generation and pip install suggestions. A reader would care because these hallucinations can be turned into supply-chain attacks when attackers register malicious packages under the fake names the model invents.

Core claim

BOUND formulates package hallucination mitigation as a package-validity boundary editing problem, where the boundary refers to the model's ability to distinguish valid packages from hallucinated package names under a given task context. It first locates modules related to package hallucination through a risk-aware localization strategy, and then edits these modules with lightweight LoRA adapters using a boundary-aware objective that reinforces valid packages, suppresses hallucinated packages, and preserves locality behavior. Experimental results show that BOUND effectively reduces package hallucinations while preserving valid package recommendations, with package-level hallucination rate red

What carries the argument

The package-validity boundary (the model's learned distinction between valid and hallucinated package names under task context), refined through risk-aware module localization followed by boundary-aware LoRA editing.

If this is right

  • Package hallucination rate falls 79.9 percent on the prompts used for editing and 65.4 percent on new prompts in package recommendation.
  • The edited boundary transfers to code generation, lowering hallucination rate by 12.8 percent.
  • The same boundary transfers to pip install recommendation, lowering hallucination rate by 34.0 percent.
  • Valid package outputs remain intact while hallucinated outputs are suppressed.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same localization-plus-boundary approach could be tested on other types of code-related hallucinations such as function-name or API errors.
  • If the boundary generalizes across model sizes, the method might reduce the need for full retraining on larger LLMs.
  • Combining the edited model with an external package registry check at inference time could create a layered defense against supply-chain attacks.

Load-bearing premise

The risk-aware localization step accurately identifies the modules that control package hallucination behavior, and the boundary-aware objective can strengthen valid packages and weaken invalid ones without harming unrelated model abilities.

What would settle it

After applying BOUND, the package hallucination rate on unseen prompts stays the same or rises, or the rate of correct package recommendations falls.

Figures

Figures reproduced from arXiv: 2607.02052 by Kui Liu, Shuhan Liu, Xiaohu Yang, Xing Hu, Xin Xia, Yukai Zhao.

Figure 1
Figure 1. Figure 1: Motivating example of package hallucination as a package [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of BOUND. ❶ (left) BOUND constructs edit cases by identifying valid packages, hallucinated packages, and building locality prompts from high-risk task prompts. ❷ (middle) BOUND computes a boundary risk score and backpropagates it through the frozen LLM to locate modules that most affect the package-validity boundary. ❸ (right) BOUND injects LoRA into the localized modules and optimizes them with c… view at source ↗
read the original abstract

Large language models (LLMs) have demonstrated strong capabilities in software engineering tasks, such as code generation, library recommendation, and dependency configuration. However, recent studies show that LLMs may suffer from package hallucination, where they generate non-existent or invalid package names. These hallucinations can be exploited in software supply chain attacks, as attackers may register malicious packages under hallucinated names. Therefore, mitigating package hallucination is important for improving the reliability and security of LLM-assisted software development. In this paper, we introduce BOUND, a lightweight localized model editing framework for mitigating package hallucinations in LLMs. BOUND formulates package hallucination mitigation as a package-validity boundary editing problem, where the boundary refers to the model's ability to distinguish valid packages from hallucinated package names under a given task context. It first locates modules related to package hallucination through a risk-aware localization strategy, and then edits these modules with lightweight LoRA adapters using a boundary-aware objective that reinforces valid packages, suppresses hallucinated packages, and preserves locality behavior. Experimental results show that BOUND effectively reduces package hallucinations while preserving valid package recommendations. In the package recommendation task, BOUND reduces package-level hallucination rate (Package-HR) by 79.9% on edit prompts and by 65.4% on unseen prompts. The learned package-validity boundary further generalizes to other package-related tasks, reducing Package-HR by 12.8% in code generation and by 34.0% in pip install recommendation. These results show that BOUND refines the package-validity boundary of LLMs and improves the reliability of package-related outputs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes BOUND, a localized model-editing method that first applies risk-aware localization to identify modules tied to package hallucination and then fine-tunes them with LoRA adapters under a boundary-aware objective (reinforce valid packages, suppress hallucinated ones, preserve locality). On a package-recommendation task the method is reported to cut Package-HR by 79.9 % on edit prompts and 65.4 % on unseen prompts; the same boundary is claimed to generalize, yielding 12.8 % and 34.0 % reductions in code-generation and pip-install settings respectively.

Significance. If the localization step truly isolates package-validity behavior and the editing objective introduces no measurable side-effects on unrelated capabilities, the work would supply a practical, lightweight intervention for a concrete security-relevant failure mode in LLM-assisted software engineering. The framing as “package-validity boundary editing” and the reported cross-task generalization are the most distinctive contributions; however, the absence of locality metrics, ablation controls, and statistical detail in the reported experiments prevents a firm judgment of impact.

major comments (3)
  1. [§4] §4 (Experimental Setup) and Table 2: the headline Package-HR reductions (79.9 % / 65.4 %) are presented without any baseline model, statistical test, error bars, or dataset-size information, so it is impossible to determine whether the gains exceed what would be expected from random variation or from a generic capability shift.
  2. [§3.2] §3.2 (Risk-aware Localization) and §3.3 (Boundary-aware Objective): no quantitative locality or side-effect metrics (e.g., accuracy on non-package tokens, performance on unrelated SE tasks, or activation-shift norms) are supplied to verify that the identified modules affect only package validity and that the objective does not alter decision boundaries for other tokens or tasks.
  3. [§4.3] §4.3 (Generalization Experiments): the reported 12.8 % and 34.0 % reductions on code generation and pip-install tasks are given without controls that isolate the contribution of the edited boundary from any incidental change in overall model behavior.
minor comments (2)
  1. [§2] Notation for Package-HR is introduced without an explicit equation; a formal definition would improve reproducibility.
  2. [§4] The abstract and §4 omit the base model size, LoRA rank, and learning-rate schedule; these details belong in the experimental protocol.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for the constructive feedback on our manuscript. We agree that strengthening the experimental reporting with additional statistical details, locality metrics, and controls will improve the rigor of the presentation. We address each major comment below and will incorporate the necessary revisions.

read point-by-point responses

  1. Referee: [§4] §4 (Experimental Setup) and Table 2: the headline Package-HR reductions (79.9 % / 65.4 %) are presented without any baseline model, statistical test, error bars, or dataset-size information, so it is impossible to determine whether the gains exceed what would be expected from random variation or from a generic capability shift.

    Authors: The reductions are computed relative to the unedited base model, which serves as the baseline. We acknowledge that the current version omits error bars, statistical tests, and explicit dataset sizes. In the revised manuscript we will add these elements, including standard deviations from repeated runs, p-values from paired statistical tests, and full details on prompt counts and dataset composition. revision: yes

  2. Referee: [§3.2] §3.2 (Risk-aware Localization) and §3.3 (Boundary-aware Objective): no quantitative locality or side-effect metrics (e.g., accuracy on non-package tokens, performance on unrelated SE tasks, or activation-shift norms) are supplied to verify that the identified modules affect only package validity and that the objective does not alter decision boundaries for other tokens or tasks.

    Authors: The manuscript describes the localization and objective but does not report numerical locality or side-effect measurements. We will add these metrics in revision, specifically token-level accuracy outside package names, performance on unrelated software-engineering benchmarks, and activation-shift norms, to confirm that edits remain localized to the package-validity boundary. revision: yes

  3. Referee: [§4.3] §4.3 (Generalization Experiments): the reported 12.8 % and 34.0 % reductions on code generation and pip-install tasks are given without controls that isolate the contribution of the edited boundary from any incidental change in overall model behavior.

    Authors: The generalization experiments demonstrate transfer of the edited boundary. To isolate its contribution we will include additional controls in the revision, such as comparison against models edited with random or unrelated objectives, thereby showing that the observed reductions stem from the package-validity boundary rather than nonspecific behavioral shifts. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method with external benchmarks

full rationale

The paper presents an empirical editing framework (BOUND) evaluated via measured reductions in Package-HR on held-out prompts and downstream tasks. No equations, derivations, or fitted parameters are invoked as predictions; results rest on direct experimental outcomes against external test sets rather than any self-referential reduction. Self-citations, if present, are not load-bearing for any claimed derivation chain. The work is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Ledger populated from abstract only; full paper unavailable so entries are minimal and provisional.

axioms (1)
  • domain assumption LoRA adapters can be used for localized behavioral edits in LLMs without global retraining
    The method depends on this established property of parameter-efficient fine-tuning.
invented entities (1)
  • package-validity boundary no independent evidence
    purpose: Conceptual distinction inside the model between valid and hallucinated package names under task context
    Introduced as the target of editing; no independent falsifiable evidence supplied in abstract.

pith-pipeline@v0.9.1-grok · 5837 in / 1337 out tokens · 31722 ms · 2026-07-03T08:51:13.433682+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

58 extracted references · 17 canonical work pages · 6 internal anchors

  1. [1]

    A review on code generation with llms: Application and evaluation,

    J. Wang and Y . Chen, “A review on code generation with llms: Application and evaluation,” in2023 IEEE International Conference on Medical Artificial Intelligence (MedAI). IEEE, 2023, pp. 284–289

    2023

  2. [2]

    Swe-bench: Can language models resolve real-world github issues?

    C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan, “Swe-bench: Can language models resolve real-world github issues?” inInternational Conference on Learning Representa- tions, vol. 2024, 2024, pp. 54 107–54 157

    2024

  3. [3]

    Code Llama: Open Foundation Models for Code

    B. Roziere, J. Gehring, F. Gloeckle, S. Sootla, I. Gat, X. E. Tan, Y . Adi, J. Liu, R. Sauvestre, T. Remezet al., “Code llama: Open foundation models for code,”arXiv preprint arXiv:2308.12950, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  4. [4]

    Source code summarization in the era of large language models,

    W. Sun, Y . Miao, Y . Li, H. Zhang, C. Fang, Y . Liu, G. Deng, Y . Liu, and Z. Chen, “Source code summarization in the era of large language models,” in2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE). IEEE, 2025, pp. 1882–1894

    2025

  5. [5]

    Depsrag: Towards agentic reasoning and planning for software dependency management,

    M. Alhanahnah and Y . Boshmaf, “Depsrag: Towards agentic reasoning and planning for software dependency management,”arXiv preprint arXiv:2405.20455, 2024

    work page arXiv 2024

  6. [6]

    How robust are llm- generated library imports? an empirical study using stack overflow,

    J. Latendresse, S. Khatoonabadi, and E. Shihab, “How robust are llm- generated library imports? an empirical study using stack overflow,” arXiv preprint arXiv:2507.10818, 2025

    work page arXiv 2025

  7. [7]

    Asleep at the keyboard? assessing the security of github copilot’s code con- tributions,

    H. Pearce, B. Ahmad, B. Tan, B. Dolan-Gavitt, and R. Karri, “Asleep at the keyboard? assessing the security of github copilot’s code con- tributions,”Communications of the ACM, vol. 68, no. 2, pp. 96–105, 2025

    2025

  8. [8]

    Do users write more insecure code with ai assistants?

    N. Perry, M. Srivastava, D. Kumar, and D. Boneh, “Do users write more insecure code with ai assistants?” inProceedings of the 2023 ACM SIGSAC conference on computer and communications security, 2023, pp. 2785–2799

    2023

  9. [9]

    Siren’s song in the ai ocean: A survey on hallucination in large language models,

    Y . Zhang, Y . Li, L. Cui, D. Cai, L. Liu, T. Fu, X. Huang, E. Zhao, Y . Zhang, Y . Chenet al., “Siren’s song in the ai ocean: A survey on hallucination in large language models,”Computational Linguistics, vol. 51, no. 4, pp. 1373–1418, 2025

    2025

  10. [10]

    The current challenges of software engineering in the era of large language models,

    C. Gao, X. Hu, S. Gao, X. Xia, and Z. Jin, “The current challenges of software engineering in the era of large language models,”ACM Transactions on Software Engineering and Methodology, vol. 34, no. 5, pp. 1–30, 2025

    2025

  11. [11]

    Llm hallucinations in practical code generation: Phenomena, mechanism, and mitigation,

    Z. Zhang, C. Wang, Y . Wang, E. Shi, Y . Ma, W. Zhong, J. Chen, M. Mao, and Z. Zheng, “Llm hallucinations in practical code generation: Phenomena, mechanism, and mitigation,”Proceedings of the ACM on Software Engineering, vol. 2, no. ISSTA, pp. 481–503, 2025

    2025

  12. [12]

    We have a package for you! a comprehensive analysis of package hallucinations by code generating LLMs,

    J. Spracklen, R. Wijewickrama, A. N. Sakib, A. Maiti, and B. Viswanath, “We have a package for you! a comprehensive analysis of package hallucinations by code generating LLMs,” in34th USENIX Security Symposium (USENIX Security 25), 2025, pp. 3687–3706

    2025

  13. [13]

    Importing phantoms: Measuring llm package hallucination vulnerabilities,

    A. Krishna, E. Galinkin, L. Derczynski, and J. Martin, “Importing phantoms: Measuring llm package hallucination vulnerabilities,”arXiv preprint arXiv:2501.19012, 2025

    work page arXiv 2025

  14. [14]

    Hfuzzer: Testing large language models for package hallucinations via phrase-based fuzzing,

    Y . Zhao, M. Wu, X. Hu, and X. Xia, “Hfuzzer: Testing large language models for package hallucinations via phrase-based fuzzing,”arXiv preprint arXiv:2509.23835, 2025

    work page arXiv 2025

  15. [15]

    Backstabber’s knife collection: A review of open source software supply chain attacks,

    M. Ohm, H. Plate, A. Sykosch, and M. Meier, “Backstabber’s knife collection: A review of open source software supply chain attacks,” in International Conference on Detection of Intrusions and Malware, and Vulnerability Assessment. Springer, 2020, pp. 23–43

    2020

  16. [16]

    Sok: Taxonomy of attacks on open-source software supply chains,

    P. Ladisa, H. Plate, M. Martinez, and O. Barais, “Sok: Taxonomy of attacks on open-source software supply chains,” in2023 IEEE Symposium on Security and Privacy (SP). IEEE, 2023, pp. 1509–1526

    2023

  17. [17]

    Beyond typosquatting: an in-depth look at package confusion,

    S. Neupane, G. Holmes, E. Wyss, D. Davidson, and L. De Carli, “Beyond typosquatting: an in-depth look at package confusion,” in32nd USENIX security symposium (USENIX security 23), 2023, pp. 3439– 3456

    2023

  18. [18]

    Diving deeper into ai package hallucinations,

    B. Lanyado, “Diving deeper into ai package hallucinations,” https://www.lasso.security/blog/ai-package-hallucinations, 2024

    2024

  19. [19]

    A survey on common threats in npm and pypi registries,

    B. Kaplan and J. Qian, “A survey on common threats in npm and pypi registries,” inInternational Workshop on Deployable Machine Learning for Security Defense. Springer, 2021, pp. 132–156

    2021

  20. [20]

    PyPI: The python package index,

    Python Software Foundation, “PyPI: The python package index,” https://pypi.org/, 2026, accessed: 2026-06-14

    2026

  21. [21]

    Retrieval- augmented generation for knowledge-intensive nlp tasks,

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. K ¨uttler, M. Lewis, W.-t. Yih, T. Rockt ¨aschelet al., “Retrieval- augmented generation for knowledge-intensive nlp tasks,”Advances in neural information processing systems, vol. 33, pp. 9459–9474, 2020

    2020

  22. [22]

    On mitigating code llm hallucinations with api documentation,

    N. Jain, R. Kwiatkowski, B. Ray, M. K. Ramanathan, and V . Kumar, “On mitigating code llm hallucinations with api documentation,” in2025 IEEE/ACM 47th International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP). IEEE, 2025, pp. 237– 248

    2025

  23. [23]

    Towards mitigating llm hallucination via self reflection,

    Z. Ji, T. Yu, Y . Xu, N. Lee, E. Ishii, and P. Fung, “Towards mitigating llm hallucination via self reflection,” inFindings of the Association for Computational Linguistics: EMNLP 2023, 2023, pp. 1827–1843

    2023

  24. [24]

    Self-refine: Iter- ative refinement with self-feedback,

    A. Madaan, N. Tandon, P. Gupta, S. Hallinan, L. Gao, S. Wiegreffe, U. Alon, N. Dziri, S. Prabhumoye, Y . Yanget al., “Self-refine: Iter- ative refinement with self-feedback,”Advances in neural information processing systems, vol. 36, pp. 46 534–46 594, 2023

    2023

  25. [25]

    A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,

    L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qinet al., “A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,” ACM Transactions on Information Systems, vol. 43, no. 2, pp. 1–55, 2025

    2025

  26. [26]

    A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Models

    S. Tonmoy, S. Zaman, V . Jain, A. Rani, V . Rawte, A. Chadha, and A. Das, “A comprehensive survey of hallucination mitigation techniques in large language models,”arXiv preprint arXiv:2401.01313, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  27. [27]

    Model editing for llms4code: How far are we?

    X. Li, S. Wang, S. Li, J. Ma, J. Yu, X. Liu, J. Wang, B. Ji, and W. Zhang, “Model editing for llms4code: How far are we?” in2025 IEEE/ACM 47th International Conference on Software Engineering (ICSE). IEEE, 2025, pp. 937–949

    2025

  28. [28]

    Detoxifying large language models via knowledge editing,

    M. Wang, N. Zhang, Z. Xu, Z. Xi, S. Deng, Y . Yao, Q. Zhang, L. Yang, J. Wang, and H. Chen, “Detoxifying large language models via knowledge editing,” inProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2024, pp. 3093–3118

    2024

  29. [29]

    A comprehensive study of knowledge editing for large language models,

    N. Zhang, Y . Yao, B. Tian, P. Wang, S. Deng, M. Wang, Z. Xi, S. Mao, J. Zhang, Y . Niet al., “A comprehensive study of knowledge editing for large language models,”arXiv preprint arXiv:2401.01286, 2024

    work page arXiv 2024

  30. [30]

    Creme: Robustness enhancement of code llms via layer-aware model editing,

    S. Liu, X. Hu, K. Huang, X. Yang, D. Lo, and X. Xia, “Creme: Robustness enhancement of code llms via layer-aware model editing,” arXiv preprint arXiv:2507.16407, 2025

    work page arXiv 2025

  31. [31]

    Can knowledge editing really correct hallucinations?

    B. Huang, C. Chen, X. Xu, A. Payani, and K. Shu, “Can knowledge editing really correct hallucinations?” inInternational Conference on Learning Representations, vol. 2025, 2025, pp. 88 116–88 149

    2025

  32. [32]

    Knowledge neurons in pretrained transformers,

    D. Dai, L. Dong, Y . Hao, Z. Sui, B. Chang, and F. Wei, “Knowledge neurons in pretrained transformers,” inProceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022, pp. 8493–8502

    2022

  33. [33]

    Pmet: Precise model editing in a transformer,

    X. Li, S. Li, S. Song, J. Yang, J. Ma, and J. Yu, “Pmet: Precise model editing in a transformer,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 17, 2024, pp. 18 564–18 572

    2024

  34. [34]

    Locating and editing factual associations in gpt,

    K. Meng, D. Bau, A. Andonian, and Y . Belinkov, “Locating and editing factual associations in gpt,”Advances in neural information processing systems, vol. 35, pp. 17 359–17 372, 2022

    2022

  35. [35]

    Depn: Detecting and editing privacy neurons in pretrained language models,

    X. Wu, J. Li, M. Xu, W. Dong, S. Wu, C. Bian, and D. Xiong, “Depn: Detecting and editing privacy neurons in pretrained language models,” inProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, 2023, pp. 2875–2886

    2023

  36. [36]

    Mass-Editing Memory in a Transformer

    K. Meng, A. S. Sharma, A. Andonian, Y . Belinkov, and D. Bau, “Mass- editing memory in a transformer,”arXiv preprint arXiv:2210.07229, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  37. [37]

    Transformer feed-forward layers are key-value memories,

    M. Geva, R. Schuster, J. Berant, and O. Levy, “Transformer feed-forward layers are key-value memories,” inProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021, pp. 5484– 5495

    2021

  38. [38]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chenet al., “Lora: Low-rank adaptation of large language models.” Iclr, vol. 1, no. 2, p. 3, 2022

    2022

  39. [39]

    Neural text generation with unlikelihood training,

    S. Welleck, I. Kulikov, S. Roller, E. Dinan, K. Cho, and J. Weston, “Neural text generation with unlikelihood training,”arXiv preprint arXiv:1908.04319, 2019

    work page arXiv 1908

  40. [40]

    DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence

    D. Guo, Q. Zhu, D. Yang, Z. Xie, K. Dong, W. Zhang, G. Chen, X. Bi, Y . Wu, Y . Liet al., “Deepseek-coder: when the large language model meets programming–the rise of code intelligence,”arXiv preprint arXiv:2401.14196, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  41. [41]

    Qwen3 Technical Report

    A. Yang, A. Li, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Gao, C. Huang, C. Lvet al., “Qwen3 technical report,”arXiv preprint arXiv:2505.09388, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  42. [42]

    The Llama 3 Herd of Models

    A. Grattafiori, A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Vaughanet al., “The llama 3 herd of models,”arXiv preprint arXiv:2407.21783, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  43. [43]

    Survey of hallucination in natural language generation,

    Z. Ji, N. Lee, R. Frieske, T. Yu, D. Su, Y . Xu, E. Ishii, Y . J. Bang, A. Madotto, and P. Fung, “Survey of hallucination in natural language generation,”ACM computing surveys, vol. 55, no. 12, pp. 1–38, 2023

    2023

  44. [44]

    Large language models hallucination: A comprehensive survey,

    A. Alansari and H. Luqman, “Large language models hallucination: A comprehensive survey,”Computer Science Review, vol. 61, p. 100970, 2026

    2026

  45. [45]

    Exploring and evaluating hallucinations in llm-powered code generation,

    F. Liu, Y . Liu, L. Shi, H. Huang, R. Wang, Z. Yang, L. Zhang, Z. Li, and Y . Ma, “Exploring and evaluating hallucinations in llm-powered code generation,”arXiv e-prints, pp. arXiv–2404, 2024

    2024

  46. [46]

    Codehalu: Investigating code hallucinations in llms via execution-based verification,

    Y . Tian, W. Yan, Q. Yang, X. Zhao, Q. Chen, W. Wang, Z. Luo, L. Ma, and D. Song, “Codehalu: Investigating code hallucinations in llms via execution-based verification,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 24, 2025, pp. 25 300–25 308

    2025

  47. [47]

    Truthx: Alleviating hallucinations by editing large language models in truthful space,

    S. Zhang, T. Yu, and Y . Feng, “Truthx: Alleviating hallucinations by editing large language models in truthful space,” inProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2024, pp. 8908–8949

    2024

  48. [48]

    Purr: Efficiently editing language model hallucinations by denoising language model corruptions,

    A. Chen, P. Pasupat, S. Singh, H. Lee, and K. Guu, “Purr: Efficiently editing language model hallucinations by denoising language model corruptions,”arXiv preprint arXiv:2305.14908, 2023

    work page arXiv 2023

  49. [49]

    Matthew Sotoudeh and A Thakur

    A. Sinitsin, V . Plokhotnyuk, D. Pyrkin, S. Popov, and A. Babenko, “Editable neural networks,”arXiv preprint arXiv:2004.00345, 2020

    work page arXiv 2004

  50. [50]

    Modifying memories in transformer models,

    C. Zhu, A. S. Rawat, M. Zaheer, S. Bhojanapalli, D. Li, F. Yu, and S. Kumar, “Modifying memories in transformer models,”arXiv preprint arXiv:2012.00363, 2020

    work page arXiv 2012

  51. [51]

    Easyedit: An easy-to-use knowledge editing framework for large language models,

    P. Wang, N. Zhang, B. Tian, Z. Xi, Y . Yao, Z. Xu, M. Wang, S. Mao, X. Wang, S. Chenget al., “Easyedit: An easy-to-use knowledge editing framework for large language models,” inProceedings of the 62nd An- nual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), 2024, pp. 82–93

    2024

  52. [52]

    Memory- based model editing at scale,

    E. Mitchell, C. Lin, A. Bosselut, C. D. Manning, and C. Finn, “Memory- based model editing at scale,” inInternational Conference on Machine Learning. PMLR, 2022, pp. 15 817–15 831

    2022

  53. [53]

    Aging with grace: Lifelong model editing with discrete key-value adaptors,

    T. Hartvigsen, S. Sankaranarayanan, H. Palangi, Y . Kim, and M. Ghas- semi, “Aging with grace: Lifelong model editing with discrete key-value adaptors,”Advances in Neural Information Processing Systems, vol. 36, pp. 47 934–47 959, 2023

    2023

  54. [54]

    Can we edit factual knowledge by in-context learning?

    C. Zheng, L. Li, Q. Dong, Y . Fan, Z. Wu, J. Xu, and B. Chang, “Can we edit factual knowledge by in-context learning?” inProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, 2023, pp. 4862–4876

    2023

  55. [55]

    Editing factual knowledge in language models,

    N. De Cao, W. Aziz, and I. Titov, “Editing factual knowledge in language models,” inProceedings of the 2021 conference on empirical methods in natural language processing, 2021, pp. 6491–6506

    2021

  56. [56]

    Fast model editing at scale,

    E. Mitchell, C. Lin, A. Bosselut, C. Finn, and C. D. Manning, “Fast model editing at scale,”arXiv preprint arXiv:2110.11309, 2021

    work page arXiv 2021

  57. [57]

    Model editing harms general abilities of large language models: Regularization to the rescue,

    J.-C. Gu, H.-X. Xu, J.-Y . Ma, P. Lu, Z.-H. Ling, K.-W. Chang, and N. Peng, “Model editing harms general abilities of large language models: Regularization to the rescue,” inProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 2024, pp. 16 801–16 819

    2024

  58. [58]

    Model editing at scale leads to gradual and catastrophic forgetting,

    A. Gupta, A. Rao, and G. Anumanchipalli, “Model editing at scale leads to gradual and catastrophic forgetting,” inFindings of the Association for Computational Linguistics: ACL 2024, 2024, pp. 15 202–15 232

    2024