Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Theoretical Foundations and Mitigation of Hallucination in Large Language Models

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Hallucination can be framed as a measurable risk with formal bounds.

desk verdict A solid survey of hallucination detection and mitigation wrapped in a theory framing that doesn't hold: the PAC-Bayes bound is standard math applied to a 0-1 loss that is only well-defined when a unique correct output exists. read the letter →

arxiv 2507.22915 v1 pith:LKNC7YAG submitted 2025-07-20 cs.CL cs.AI

Esmail Gumaan This is my paper · ORCID
classification cs.CLcs.AI MSC 68T5068Q3268T07
keywords LLMhallucinationriskPAC-BayesboundRademachercomplexityretrieval-augmentedgenerationfactualityevaluationdetectionmitigationworkflow
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to put LLM hallucination on a formal footing by defining intrinsic hallucinations (output contradicts the source) and extrinsic hallucinations (output adds unverifiable content), then defining a hallucination risk as the expected probability that a model's output is unfaithful to a ground-truth oracle. Its main theoretical claim is that this risk is bounded above by the empirical hallucination rate plus a complexity penalty, via PAC-Bayes and Rademacher arguments, so that a low observed hallucination rate on a finite sample gives a high-probability guarantee about future inputs. The paper also argues, citing an impossibility result, that completely eliminating hallucination from a general-purpose model is not feasible, which reframes the goal as reducing and managing risk. The rest of the work surveys detection techniques, mitigation techniques, and evaluation protocols, and proposes a generate-detect-mitigate workflow. A sympathetic reader would care because this turns an elusive failure mode into a quantity one can measure, bound, and optimize.

What carries the argument

The load-bearing object is the hallucination indicator $H(M,x)$, which returns 1 when the output $y=M(x)$ is not completely faithful to the oracle $f^*(x)$, and the hallucination risk $R_{\mathrm{hall}}(M)=\mathbb{E}_x[H(M,x)]$ built from it. The argument is carried by applying the PAC-Bayes generalization bound for 0-1 loss to this indicator: the expected risk of a model distribution $Q$ is bounded by its empirical hallucination rate plus a term depending on the KL divergence between $Q$ and a prior $P$ and on the sample size. The same indicator permits Rademacher and VC-dimension complexity bounds. A separate impossibility argument for computable oracles supplies the negative result that some hallucinations are unavoidable, which is what motivates the paper's detection and mitigation workflow rather than a search for a perfectly truthful model.

What would settle it

Take a set of open-ended or subjective prompts, have independent annotators label each model output as hallucinated or faithful, and measure inter-annotator agreement. If agreement is near chance, the 0-1 oracle assumption fails for that input distribution, and the PAC-Bayes bound does not apply. A more direct check: on a distribution with a well-defined oracle, compute the bound's right-hand side from a training sample and compare it with the observed hallucination rate on a held-out sample; if the held-out rate systematically exceeds the bound, the claimed inequality would need re-examination.

Watch

Extended reading notes

Core claim

The central claim is that, under the assumption of a ground-truth oracle $f^*(x)$ that labels each input with a fully truthful output, hallucination becomes a binary 0-1 classification error, and the model's expected hallucination risk $R_{\mathrm{hall}}(M)=\mathbb{E}_x[H(M,x)]$ can be controlled by sample-based bounds. The paper derives the PAC-Bayes bound $\mathbb{E}_{M\sim Q}[R_{\mathrm{hall}}(M)] \le \mathbb{E}_{M\sim Q}[\hat{R}_{\mathrm{hall}}(M)] + \sqrt{\frac{\mathrm{KL}(Q\parallel P)+\ln(1/\delta)}{2n}}$, holding with probability at least $1-\delta$, and it invokes Rademacher/VC uniform-convergence bounds of the form $R_{\mathrm{hall}}(M) \le \hat{R}_{\mathrm{hall}}(M)+O(\sqrt{C/n})$. It further claims that for sufficiently powerful computable models no perfect reproduction of an arbitrary oracle is possible, so hallucinations are unavoidable in full generality. The paper presents these results as the theoretical foundation for the practical program: measure hallucination rates, bound them, detect likely errors, and mitigate them with retrieval, fine-tuning, calibrated decoding, and verification.

Load-bearing premise

The entire bound depends on there being a fixed ground-truth oracle that can label every output as hallucinated or faithful; for open-ended, subjective, or creative tasks no such oracle exists, so the empirical hallucination rate the bound starts from is undefined.

Editorial extensions

If this is right

  • A model with a low hallucination rate on a representative sample, and with a small KL penalty relative to its prior, comes with a formal guarantee that its true hallucination risk is also low, provided the inputs have a well-defined oracle.
  • Because the complexity term shrinks as $1/\sqrt{n}$, collecting more labeled samples tightens the guarantee on hallucination risk at a known rate.
  • Fine-tuning on high-quality factual responses and keeping the model close to its pre-fine-tuning prior directly reduces the KL term, improving the bound.
  • The claimed impossibility of perfect truthfulness implies that retrieval, verification, and refusal are not optional extras but necessary components of any reliable deployment.
  • The proposed workflow formalizes the practical consequence: generate a draft, check it with uncertainty, factuality, and source-alignment signals, and then retrieve, correct, or abstain before final delivery.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The oracle assumption suggests the formal bounds apply cleanly to closed tasks with verifiable answers, not to creative or subjective generation; extending them would require replacing the 0-1 oracle with graded human judgments, which would change the bound's meaning.
  • A natural next step the paper leaves implicit is to measure the bound's looseness on real LLMs by reporting the empirical rate, the KL term, and the held-out rate together; if the gap is large, data-dependent priors or tighter PAC-Bayes bounds could make the guarantee practically useful.
  • The generate-detect-mitigate loop can be tested as an engineering tradeoff: end-to-end hallucination rate versus latency and retrieval cost, with an optimal number of verification loops.
  • If the impossibility claim is right, retrieval-augmented generation does not remove hallucination but shifts it to cases where retrieval is incomplete or misread, so benchmarks should include unanswerable and out-of-knowledge queries.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes formal definitions of intrinsic and extrinsic hallucination, introduces a hallucination risk R_hall(M) defined through a ground-truth oracle f*(x), and states PAC-Bayes and Rademacher-complexity bounds on that risk (Sections II and III). It then surveys detection methods (uncertainty estimation, calibration, attention alignment), mitigation strategies (RAG, fine-tuning, logit calibration, fact-verification), proposes a combined detection/mitigation workflow (Section VI), and recommends evaluation datasets and metrics (Section VII). The manuscript reports no experiments and no new mathematical proof; its main formal contribution is Eq. (1), presented as a direct application of McAllester's PAC-Bayes inequality.

Significance. The paper covers a broad and important topic and its survey parts cite relevant literature accurately (e.g., semantic entropy [6], FactCC [24], TruthfulQA [28], chain-of-verification [34]). The proposed workflow is coherent and could serve as a practical checklist for system builders. However, the theoretical contribution is not novel: Eq. (1) is a standard PAC-Bayes inequality, and the central definition of hallucination risk via a unique oracle is not valid for the open-ended generation tasks that the paper explicitly targets. Because the formal object that the bound is supposed to constrain is not well-defined, the claimed theoretical foundation does not currently support the paper's conclusions.

major comments (3)
  1. [Section II.B and Section III.B, Eq. (1)] The definition of H(M,x) relies on a single ground-truth oracle f*(x) that produces one 'fully truthful and contextually appropriate' output for each input, but Section II.A states that whether content counts as a hallucination 'can depend on the task context and the expected scope of the output.' For summarization, knowledge-grounded dialogue, and open-ended QA, there are typically many fully faithful outputs (paraphrases, equally valid summaries, different correct answers), so equality to a unique f*(x) cannot define a binary hallucination indicator. The empirical rate \v{R}_hall(M) in Section III.B is therefore convention-dependent: different choices of f* change the value on the right-hand side of Eq. (1) without changing the factual quality of M. The PAC-Bayes inequality may be valid for a fixed labeling oracle, but it does not bound an objective hallucination risk for the tasks the paper targets. This is load-bearing because Eq. (1) is the paper's central theoretical result.
  2. [Section III.B, Eq. (1)] The paper says it derives a PAC-Bayes bound, but Eq. (1) is quoted from McAllester [38] as a 'direct application' with no derivation and no precise statement of the underlying supervised protocol. In particular, no labeled training sample of pairs (x_i, y_i) is defined: \v{R}_hall(M) is an average of H(M,x_i) over inputs x_i with labels supplied by an external oracle, whereas the PAC-Bayes theorem concerns empirical risk on a drawn sample with a fixed loss l(h,z) for z=(x,y). As written, the bound does not formally follow from the cited theorem, and the paper should either prove the instantiation or state the exact learning protocol (including what data is drawn, what the labels are, and over what randomness the probability 1-δ is taken).
  3. [Section III.C] The impossibility claim that hallucinations are fundamentally unavoidable is presented only as an informal paraphrase of Xu et al. [11] ('a recent theoretical result suggests...'). No theorem statement, assumptions, or proof is given, so the paper's 'theoretical takeaway' is not self-contained. If this claim is part of the paper's contribution, it should be stated as a precise theorem (e.g., in terms of computability, the distribution over inputs, and the hypothesis class) and proved or explicitly attributed as an external result outside the scope of the paper's own derivation.
minor comments (5)
  1. [Section IV.B] The phrase 'similar to how Fact was a BERT-based classifier' appears to refer to FactCC [24]; the typo should be corrected and the reference made explicit.
  2. [Section VI and Figure 1] Figure 1 is rendered as ASCII art in the manuscript; a proper figure with clearly labeled components (input, LLM, detection module, mitigation module, final answer) should be provided, since the text references it as 'Figure 1.'
  3. [Section III.A] The bound R_hall(M) ≤ \v{R}_hall(M) + O(√(C/n)) is stated with an unspecified complexity measure and unspecified constants; for Rademacher complexity the standard form is R_n + √(ln(1/δ)/(2n)) (or similar), so the exact theorem and assumptions should be cited to make the statement checkable.
  4. [Introduction, RAG description] Reference [30] is cited alongside [17] and [18] for retrieval-augmented generation, but [30] is a knowledge-based dialogue metrics paper; the citation appears misplaced and should be moved or replaced.
  5. [Section VII.C] The experimental recommendations (e.g., 'use sufficiently large sample sizes and statistical tests') are sensible, but the paper itself presents no empirical evaluation; the text should explicitly state that these are recommendations for future work, not results of this paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Eq. (1) is a cited standard PAC-Bayes bound applied to independently defined hallucination risk, not a quantity defined by the bound itself.

full rationale

The paper's only formal derivation is Eq. (1) in Section III.B, which it explicitly describes as 'a direct application of the PAC-Bayesian generalization bound for 0-1 loss' and cites McAllester [38]. The quantities Rhall(M) and the empirical hallucination rate are defined independently in Section II.B and Section III.A, respectively, and Eq. (1) relates them via a standard complexity term; neither is constructed from the other. No parameter is fitted to data and then renamed as a prediction, and the paper contains no experimental or empirical claims that could reduce to fitted inputs. The load-bearing citations are external and independently established: McAllester's PAC-Bayes theorem, Bartlett-Mendelson/Vapnik complexity bounds, and Xu et al. on the inevitability of hallucination; none of these is a self-citation by the paper's author. The oracle f*(x) assumption in Section II.B is a modeling choice that may limit applicability in open-ended or subjective tasks, but it does not make the bound equivalent to its inputs: H and Rhall are defined through f*, while the theorem is an imported inequality relating a population expectation to a sample average. Concerns about the empirical hallucination rate being convention-dependent are correctness and scope risks, not circularity. Therefore the derivation chain is not circular.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central theoretical result rests on standard learning-theory results cited from the literature, and on an idealized oracle assumption that is not justified for real-world LLM tasks.

assumptions (4)
  • domain assumption Existence of a ground-truth oracle f*(x) that provides fully truthful outputs
    Introduced in Section II.B (Hallucination Risk) and used to define H(M,x) as a binary label. Fails for open-ended or creative tasks.
  • domain assumption Hallucination can be treated as a 0-1 classification loss
    Used in Section III.B to apply McAllester's PAC-Bayes bound. Ignores gradations and subjectivity in hallucination.
  • standard math McAllester's PAC-Bayes bound for 0-1 loss holds as stated
    Invoked in Eq. (1) via reference [38], but no proof or derivation is given in the paper.
  • ad hoc to paper The impossibility result of Xu et al. [11] applies to LLMs as general problem solvers
    Section III.C leans on this cited result to claim hallucination cannot be fully eliminated, without presenting the argument.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Theoretical Foundations and Mitigation of Hallucination in Large Language Models." pith.science (2026). https://pith.science/paper/LKNC7YAG

@misc{pith2026250722915,
  author       = {Pith},
  title        = {Pith review of: Theoretical Foundations and Mitigation of Hallucination in Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LKNC7YAG}},
  note         = {Machine review of arXiv:2507.22915}
}
read the original abstract

Hallucination in Large Language Models (LLMs) refers to the generation of content that is not faithful to the input or the real-world facts. This paper provides a rigorous treatment of hallucination in LLMs, including formal definitions and theoretical analyses. We distinguish between intrinsic and extrinsic hallucinations, and define a \textit{hallucination risk} for models. We derive bounds on this risk using learning-theoretic frameworks (PAC-Bayes and Rademacher complexity). We then survey detection strategies for hallucinations, such as token-level uncertainty estimation, confidence calibration, and attention alignment checks. On the mitigation side, we discuss approaches including retrieval-augmented generation, hallucination-aware fine-tuning, logit calibration, and the incorporation of fact-verification modules. We propose a unified detection and mitigation workflow, illustrated with a diagram, to integrate these strategies. Finally, we outline evaluation protocols for hallucination, recommending datasets, metrics, and experimental setups to quantify and reduce hallucinations. Our work lays a theoretical foundation and practical guidelines for addressing the crucial challenge of hallucination in LLMs.

Figures

Figures reproduced from arXiv: 2507.22915 by the authors.

Figure 1
Figure 1. Proposed workflow for hallucination detection and mitigation. The LLM generates a draft response given an input query. A detection module then [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Hallucination is a Consequence of Space-Optimality: A Rate-Distortion Theorem for Membership Testing

    cs.LG 2026-01 unverdicted novelty 6.0 of 10

    In a closed world of sparse random facts, the minimal memory per fact equals the KL divergence between fact and non-fact score distributions, so the optimal limited-capacity strategy hallucinates a fixed fraction of n...

  2. Explainable Knowledge Graph Retrieval-Augmented Generation (KG-RAG) with KG-SMILE

    cs.AI 2025-09 reject novelty 4.0 of 10

    KG-SMILE applies perturbation and linear regression to a knowledge graph to attribute which entities and relations drive a GraphRAG system's answers.

Reference graph

Works this paper leans on

38 extracted references · 35 canonical work pages · cited by 2 Pith papers

  1. [6]

    Detecting hallucinations in large language models using semantic entropy,

    S. Farquhar, J. Kossen, L. Kuhn, and Y . Gal, “Detecting hallucinations in large language models using semantic entropy,” Nature Machine Intelligence, vol. 6, no. 9, pp. 1019–1027, 2024

    work page 2024

  2. [24]

    Evaluating the factual consistency of abstractive text summarization,

    W. Kry ´sci´nski, B. McCann, C. Xiong, and R. Socher, “Evaluating the factual consistency of abstractive text summarization,” in Proc. 2020 Conference on Empirical Methods in NLP (EMNLP) , 2020, pp. 9332– 9346

    work page 2020

  3. [28]

    TruthfulQA: Measuring how models mimic human falsehoods,

    S. Lin, J. Hilton, and O. Evans, “TruthfulQA: Measuring how models mimic human falsehoods,” in Proc. 39th International Conference on Machine Learning (ICML) , 2022, pp. 13667–13690

    work page 2022

  4. [34]

    Chain-of-verification reduces hallucination in large language models,

    S. Dhuliawala, M. Komeili, J. Tang et al., “Chain-of-verification reduces hallucination in large language models,” arXiv:2309.11495, 2023

    arXiv 2023

  5. [38]

    PAC-Bayesian model averaging,

    D. McAllester, “PAC-Bayesian model averaging,” in Proc. 12th Annual Conference on Computational Learning Theory (COLT), 1999, pp. 164– 170

    work page 1999

  6. [11]

    Layer-Adapted Implicit Distribution Alignment Networks for Cross-Corpus Speech Emotion Recognition

    Y . Xu, L. Fan, and M. Kankanhalli, “Hallucination is inevitable: An innate limitation of large language models,” arXiv:2310.03992, 2024

    work page Pith review arXiv 2024

  7. [1]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar et al., “Attention is all you need,” in Proc. Advances in Neural Information Processing Systems (NeurIPS) , 2017, pp. 5998–6008

    work page 2017

  8. [2]

    BERT: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” in Proc. 2019 Conference of the North American Chapter of the ACL (NAACL) , 2019, pp. 4171–4186

    work page 2019

Show all 38 references
  1. [3]

    Language models are few-shot learners,

    T. Brown, B. Mann, N. Ryder et al. , “Language models are few-shot learners,” in Proc. Advances in Neural Information Processing Systems (NeurIPS), 2020, pp. 1877–1901

    2020

  2. [4]

    Survey of hallucination in natural language generation,

    Z. Ji, N. Lee, R. Frieske et al. , “Survey of hallucination in natural language generation,” ACM Computing Surveys, vol. 55, no. 12, p. 248, 2023

    2023

  3. [5]

    On faithfulness and factuality in abstractive summarization,

    J. Maynez, S. Narayan, B. Bohnet, and R. McDonald, “On faithfulness and factuality in abstractive summarization,” in Proc. 58th Annual Meeting of the ACL , 2020, pp. 1906–1919

    2020

  4. [7]

    A neural conversational model,

    O. Vinyals and Q. Le, “A neural conversational model,” in Proc. ICML Deep Learning Workshop, 2015, arXiv:1506.05869

    2015 arXiv

  5. [8]

    Six challenges for neural machine transla- tion,

    P. Koehn and R. Knowles, “Six challenges for neural machine transla- tion,” in Proc. 1st Workshop on Neural Machine Translation , 2017, pp. 28–39

    2017

  6. [9]

    The curious case of hallucinations in neural machine translation,

    V . Raunak, A. Menezes, and M. Junczys-Dowmunt, “The curious case of hallucinations in neural machine translation,” in Proc. 18th International Conference on Spoken Language Translation (IWSLT) , 2021, pp. 1172– 1183

    2021

  7. [10]

    A survey on hallucination in large language models,

    Y . Zhang, A. Elgohary, M. Zhao et al. , “A survey on hallucination in large language models,” arXiv:2305.11717, 2023

    2023 arXiv

  8. [12]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning,

    Y . Gal and Z. Ghahramani, “Dropout as a bayesian approximation: Representing model uncertainty in deep learning,” in Proc. 33rd Interna- tional Conference on Machine Learning (ICML) , 2016, pp. 1050–1059

    2016

  9. [13]

    Language models (mostly) know what they know,

    N. Kadavath, S. Bai, A. Coner et al. , “Language models (mostly) know what they know,” in Proc. Neural Information Processing Systems (NeurIPS), 2022

    2022

  10. [14]

    On calibration of modern neural networks,

    C. Guo, G. Pleiss, Y . Sun, and K. Q. Weinberger, “On calibration of modern neural networks,” in Proc. 34th International Conference on Machine Learning (ICML) , 2017, pp. 1321–1330

    2017

  11. [15]

    How can we know when language models know? On the calibration of language models for question answering,

    H. Jiang, B. Kassim, H. Nepal, and M. Bansal, “How can we know when language models know? On the calibration of language models for question answering,” Transactions of the Association for Computational Linguistics, vol. 9, pp. 962–977, 2021

    2021

  12. [16]

    Get to the point: Summarization with pointer-generator networks,

    A. See, P. J. Liu, and C. D. Manning, “Get to the point: Summarization with pointer-generator networks,” in Proc. 55th Annual Meeting of the ACL, 2017, pp. 1073–1083

    2017

  13. [17]

    Retrieval-augmented generation for knowledge-intensive NLP,

    P. Lewis, E. Perez, A. Piktus et al. , “Retrieval-augmented generation for knowledge-intensive NLP,” in Proc. Advances in Neural Information Processing Systems (NeurIPS) , 2020, pp. 9459–9474

    2020

  14. [18]

    REALM: Retrieval-augmented language model pre-training,

    K. Guu, K. Lee, Z. Tong, P. Pasupat, and M. Chang, “REALM: Retrieval-augmented language model pre-training,” in Proc. 37th Inter- national Conference on Machine Learning (ICML) , 2020, pp. 3929– 3938

    2020

  15. [19]

    Knowledge-informed dialogue generation,

    K. Shuster, S. Bhuwan, M. Chen et al., “Knowledge-informed dialogue generation,” in Proc. 2021 Conference on Empirical Methods in NLP (EMNLP), 2021, pp. 1656–1679

    2021

  16. [20]

    Training language models to follow instructions with human feedback,

    L. Ouyang, J. Wu, X. Jiang et al. , “Training language models to follow instructions with human feedback,” in Proc. Advances in Neural Information Processing Systems (NeurIPS) , 2022

    2022

  17. [21]

    WebGPT: Browser-assisted question-answering with human feedback,

    R. Nakano, J. Hilton, S. Balaji et al. , “WebGPT: Browser-assisted question-answering with human feedback,” arXiv:2112.09332, 2022

    2022 arXiv

  18. [22]

    Calibration of pre-trained transformers,

    S. Desai and G. Durrett, “Calibration of pre-trained transformers,” in Proc. 2020 Conference on Empirical Methods in NLP (EMNLP) , 2020, pp. 295–302

    2020

  19. [23]

    The curious case of neural text degeneration,

    A. Holtzman, J. Buys, L. Du, M. Forbes, and Y . Choi, “The curious case of neural text degeneration,” in Proc. 8th International Conference on Learning Representations (ICLR) , 2020

    2020

  20. [25]

    SelfCheckGPT: Zero-resource black-box hallucination detection for generative large language models,

    P. Manakul, J. D’Haro, and S. Sakti, “SelfCheckGPT: Zero-resource black-box hallucination detection for generative large language models,” in Proc. 61st Annual Meeting of the ACL , 2023, pp. 2229–2244

    2023

  21. [26]

    Asking and answering questions to evaluate the factual consistency of summaries,

    X. Wang, P. Gan, Y . Zhang, and J. Zhao, “Asking and answering questions to evaluate the factual consistency of summaries,” in Proc. 58th Annual Meeting of the ACL , 2020, pp. 5008–5020

    2020

  22. [27]

    TRUE: Re-evaluating factual consistency evaluation,

    O. Honovich, E. Lahav, A. Katz, and J. Goldberg, “TRUE: Re-evaluating factual consistency evaluation,” in Proc. 2nd Workshop on Document- grounded Dialogue and Conversational QA (DialDoc@ACL) , 2022, pp. 161–175

    2022

  23. [29]

    Understanding fac- tuality in abstractive summarization with FRANK: A benchmark for factuality metrics,

    A. Pagnoni, S. Balachandran, and A. Tsvetkov, “Understanding fac- tuality in abstractive summarization with FRANK: A benchmark for factuality metrics,” in Proc. 2021 Conference on Empirical Methods in NLP (EMNLP), 2021, pp. 4818–4837

    2021

  24. [30]

    Knowledge-based metrics for dialogue: Unifying structured and unstructured knowledge,

    K. Shuster, M. Chen, S. Ju et al. , “Knowledge-based metrics for dialogue: Unifying structured and unstructured knowledge,” in Proc. 1st Workshop on Natural Language Generation, Evaluation, and Metrics (GEM@ACL), 2021, pp. 9–23

    2021

  25. [31]

    FEVER: a large-scale dataset for fact extraction and verification,

    J. Thorne, A. Vlachos, C. Christodoulopoulos, and A. Mittal, “FEVER: a large-scale dataset for fact extraction and verification,” in Proc. 2018 Conference of the North American Chapter of the ACL (NAACL) , 2018, pp. 809–819

    2018

  26. [32]

    What uncertainties do we need in bayesian deep learning for computer vision?

    A. Kendall and Y . Gal, “What uncertainties do we need in bayesian deep learning for computer vision?” in Proc. Advances in Neural Information Processing Systems (NIPS) , 2017, pp. 5574–5584

    2017

  27. [33]

    Selective question answering under domain shift,

    A. Kamath, P. De Cao, and R. Cipolla, “Selective question answering under domain shift,” in Proc. 58th Annual Meeting of the ACL , 2020, pp. 5684–5696

    2020

  28. [35]

    Mitigating hallucination by integrating knowledge graphs into LLMs,

    H. Zhao, J. Sun, G. Lai et al., “Mitigating hallucination by integrating knowledge graphs into LLMs,” in Proc. 37th AAAI Conference on Artificial Intelligence, 2023, pp. 12088–12096

    2023

  29. [36]

    Rademacher and gaussian complexi- ties: Risk bounds and structural results,

    P. L. Bartlett and S. Mendelson, “Rademacher and gaussian complexi- ties: Risk bounds and structural results,” Journal of Machine Learning Research, vol. 3, pp. 463–482, 2002

    2002

  30. [37]

    Vapnik, Statistical Learning Theory

    V . Vapnik, Statistical Learning Theory . Wiley, 1998

    1998

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.