Pith. sign in

REVIEW 3 major objections 4 minor 48 references

Learning Auxiliary Tasks Improves Reference-Free Hallucination Detection in Open-Domain Long-Form Generation

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A reference-free hallucination detector for long-form answers improves by jointly training on an auxiliary question-answering task, gaining about 3 balanced-accuracy points over fine-tuning on LongFact.

desk verdict A well-controlled empirical recipe that buys a consistent ~3-point BAcc gain over fine-tuning for reference-free hallucination detection, though the shared search-based annotation pipeline for training and test labels means part of the gain may reflect learning the annotator's biases. read the letter →

arxiv 2505.12265 v1 pith:4N6V5UML submitted 2025-05-18 cs.CL

Chengwei Qin , Wenxuan Zhou , Karthik Abinav Sankararaman , Nanshu Wang , Tengyu Xu , Alexander Radovic , Eryk Helenowski , Arya Talebzadeh
show 5 more authors
Aditya Tayade Sinong Wang Shafiq Joty Han Fang Hao Ma
This is my paper · ORCID
classification cs.CL
keywords hallucinationdetectionlong-formgenerationreference-freefine-tuningauxiliarytasklearningrationaleaugmentationinternalstatesbalancedaccuracy
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

This paper asks whether a language model can detect its own factual errors in open-domain long-form answers without relying on external fact-checking tools. The authors first show that internal confidence signals—output probabilities and entropies, including entity-focused variants—cannot separate factual from hallucinated claims better than random chance. After comparing prompting, probing, and fine-tuning, they find fine-tuning to be the strongest existing approach. Their central claim is that fine-tuning is not the ceiling: augmenting it with rationale-augmented labels and an auxiliary question-answering task (a method they call RATE-FT) raises balanced accuracy from 76.1 to 79.6 on LongFact and from 78.2 to 80.9 on biography generation, with consistent gains across four model families. If this is right, reference-free hallucination detection in long-form generation can be improved substantially by multi-task supervision alone.

What carries the argument

The load-bearing mechanism is RATE-FT (Rationale and Auxiliary Task Enhanced Fine-Tuning), a training protocol that pairs each (claim, label) example with a 'label-rationale' format—the model learns to emit True or False first and then explain—and with a generated question-answering example in which the model must answer a question about the key information in the claim. For hallucinated claims, the rationale is used to construct the correct answer, so the QA task always teaches a factual response. This second task is what carries the gain: ablations show that dropping either the rationales or the QA task noticeably lowers balanced accuracy, and controls with paraphrased claims or halved training data show the gain is not just a data-volume effect. The label-first ordering keeps inference cost identical to plain fine-tuning, since the label is read from the first output token.

What would settle it

Take a random sample of claims labeled by the paper's search-and-reason pipeline and have independent human fact-checkers label them, then retrain plain fine-tuning and RATE-FT on human-verified labels and compare on a human-verified test set; if RATE-FT's advantage over fine-tuning shrinks or disappears, the central claim is not established.

Watch

Extended reading notes

Core claim

The paper's discovery is that a detector's judgment about a claim gets more reliable when, during fine-tuning, it is asked not only to label the claim True or False but also to reconstruct the factual core of the claim through a derived question-and-answer task. The authors argue that internal states such as token probability and entropy are not informative for long claims because they reflect confidence in surface wording rather than correctness of content. They then show that fine-tuning outperforms prompting and probing, and that their RATE-FT protocol—training jointly on claim-label pairs, label-first rationales, and QA examples built from each claim—outperforms ordinary fine-tuning by about three balanced-accuracy points on LongFact. The improvement is attributed to the auxiliary QA task rather than to having more training data, because a paraphrase-augmented control with the same data volume does not match it, and halving RATE-FT's data still exceeds fine-tuning. The same pattern holds on a biography benchmark and across models of different families and sizes.

Load-bearing premise

The load-bearing premise is that the automated pipeline which decomposes long responses into claims, searches the web, and reasons about whether search results support each claim produces correct factual-versus-hallucinated labels for both training and evaluation; if that labeling is biased, the reported gains may reflect fitting the annotation style rather than genuine hallucination detection ability.

Editorial extensions

If this is right

  • Reference-free hallucination detectors for long-form answers can be trained without any external fact-checking tool at inference time, with accuracy near that of search-based verification.
  • The RATE-FT recipe transfers across model families and scales, so the gain is not tied to one backbone.
  • The auxiliary QA task, not extra training examples, drives the improvement; future detector training can focus on task design rather than data volume.
  • Adding an 'unknown' output and routing uncertain claims to external tools yields further gains, enabling hybrid pipelines that spend search budget only where the model is unsure.
  • Because RATE-FT performs better than fine-tuning across short, medium, and very long responses, the benefit persists even as response length grows.

Reading between the lines

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

  • Beyond the paper's experiments, the QA-auxiliary training can be read as teaching the model to 'repair' hallucinated claims into correct answers; a testable extension is whether RATE-FT also improves the factuality of the model's own generation, not just its detection accuracy.
  • If the search-and-reason labeling pipeline carries a systematic bias—for instance, favoring fluent claims—then part of the measured gain might be adaptation to that annotation style; rerunning on human-verified labels would determine how much of the +3 points is genuine detection skill.
  • The same multi-task recipe could apply to other fine-grained classification problems where labels are expensive but a correct-answer reconstruction is cheap, such as faithfulness hallucination detection or factual consistency scoring.
  • The uncertainty-routing result suggests a natural deployment pattern: run RATE-FT as a cheap pre-filter and reserve external verification for the uncertain minority of claims, which the paper did not explicitly propose as a system.
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 / 4 minor

Summary. The paper studies reference-free hallucination detection for open-domain long-form generation. It first reports that internal states (token probabilities and entropies) are not better than random at separating factual from hallucinated atomized claims, then compares prompting, probing, and fine-tuning, finding fine-tuning most effective. The main contribution is RATE-FT, which augments fine-tuning with (a) rationales in a 'label-rationale' format and (b) an auxiliary question-answering task built from the same claims. Experiments on LongFact and biography generation across several LLMs show consistent BAcc gains, e.g., +3 points over fine-tuning on LongFact, with ablations suggesting the gain comes from the auxiliary QA task rather than extra data.

Significance. If the results hold, RATE-FT is a practical, training-only enhancement that improves reference-free hallucination detectors without raising inference cost, because the model emits the label token first and stops after the first token. The paper also offers a systematic comparison of internal-state, prompting, probing, and fine-tuning baselines, which is useful for the community. The ablation design is a strength: Fine-Tuning_para and RATE-FT_half (Tables 4 and 5) directly address the 'more data vs. auxiliary task' confound, and the gains replicate across four model families and two datasets. The main weaknesses are the lack of any manual audit of the automatically constructed labels and the lack of clarity about the train/test split granularity, both of which are load-bearing for the central claim. Code and data are not released, limiting reproducibility.

major comments (3)
  1. [Section 3 and Section 5.1.6] The same automated pipeline (LLM-based claim decomposition, multi-step Google Search queries, and LLM reasoning over search results) is used to create the training, validation, and test labels, and Section 5.1.6 confirms that external search is used for all supervision. No manual audit, inter-annotator agreement, or human-verified subset is reported. Because RATE-FT additionally trains on rationales and QA examples generated from these same labels, the reported improvement over Fine-Tuning may reflect learning the annotator's systematic biases (e.g., accepting fluent but wrong claims) rather than improved true hallucination detection. The paper should provide a manual quality audit of a random sample of labels, or evaluate RATE-FT on a human-verified benchmark, to establish that the gains are not an artifact of the shared annotation source. This is essential for the central claim that RATE-FT improves hallucination detection.
  2. [Section 4] The 2,711 claims are split into training (70%), validation (20%), and test (10%) sets, but the paper does not state whether the split is at the claim level or grouped by the original response/prompt. If claims from the same long-form response appear in both training and test sets, the reported BAcc numbers, including the +3% gain of RATE-FT, may be inflated by response-level leakage, because the model could memorize response-specific patterns. The authors should clarify the split granularity; if it is claim-level, they should re-run the main comparisons with a response-level split and report whether the gains persist.
  3. [Table 3] The statement 'RATE-FT is significantly better than Fine-Tuning with p-value<0.01' is not supported by any described statistical test. A single BAcc number per method on one test set cannot by itself yield a p-value; the authors must specify the test (e.g., paired bootstrap over claims, multiple random seeds, or a significance test across models/datasets) and report the corresponding test statistic and confidence interval. This is directly relevant to the central claim of a significant improvement.
minor comments (4)
  1. [Section 3] The conclusion that internal states are 'not better than random guessing' is based only on visual inspection of overlapping histograms in Figures 2--5. Please also report quantitative metrics such as AUC or BAcc with confidence intervals to support this claim.
  2. [Section 5.1.4] The BAcc-unknown metric counts 'unknown' predictions as correct, assuming the external tool is perfect. This makes the metric sensitive to the hyperparameters α_low and α_high and can favor methods that output more unknowns. The paper should report the fraction of 'unknown' predictions and consider a metric that accounts for the cost of tool invocation.
  3. [Section 5.1.3] The comparison with F2 (Hu et al., 2024) is only one sentence and does not include any experimental comparison. Please either add a quantitative comparison or clarify in what sense the methods differ beyond the stated goal (detection vs. mitigation).
  4. [General] The paper does not release data or code, which limits reproducibility and makes external verification of label quality impossible. Releasing the constructed dataset and training scripts would strengthen the paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the claimed gains are empirical benchmark comparisons; shared search-based labels are a validity concern, not a circular reduction.

full rationale

No circular derivation chain is present. The paper is an empirical study: Section 3 measures internal states against claims labeled by LLM decomposition plus Google Search; Section 4 compares prompting, probing, and fine-tuning on a train/validation/test split of those labels; Section 5 introduces RATE-FT by augmenting fine-tuning with rationale- and QA-formatted versions of the same training claims. The central claim, that RATE-FT improves balanced accuracy over fine-tuning (e.g., 79.6 vs 76.1 on LongFact, Table 3), is an empirical comparison on held-out test labels, not a quantity that equals its input by construction. The closest concern is that the same automated Google-Search annotation pipeline produces the training, validation, and test labels (Section 3; Section 5.1.6 states the search is used only offline for constructing the benchmark). That is a potential label-bias/validity threat to generalization to true hallucination, but it is not circularity in the derivation sense: the reported BAcc values are not fitted parameters renamed as predictions, and no self-citation chain or uniqueness theorem forces the result. The paper itself acknowledges the external-search dependence in Section 5.1.6 and confines its claim to reference-free inference. There are no load-bearing self-citations (the only author-overlapping reference, Laban et al. 2023, is background in Related Work), and the comparison against F2 (Hu et al. 2024) and the Fine-Tuningpara / RATE-FThalf ablations (Tables 4-5) provide independent control conditions. Therefore no circular step can be exhibited.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new theoretical entities, particles, forces, or conserved quantities. Its central claim rests on empirical data construction: external-search-derived labels, model-generated rationales, and model-generated QA examples. The main unverified inputs are the quality of the annotation pipeline and the small-scale evaluation, both listed above.

free parameters (2)
  • Pfactual classification threshold = Optimal threshold on validation set
    The probability output by PromptTF, probing, fine-tuning, and RATE-FT is converted to a factual or hallucinated label by thresholding; the threshold is selected by a validation search, so it is a fitted degree of freedom, though not fit on the test set.
  • alpha_low and alpha_high uncertainty thresholds = Validation-optimized with BAcc above 70% constraint
    Used in the hybrid detection pipeline of Section 5.1.4 to route ambiguous claims to external tools; the optimal values are chosen by validation search, which affects the reported BAcc-unknown results.
assumptions (3)
  • domain assumption Google Search-based verification produces reliable ground-truth labels for factual and hallucinated claims.
    All training labels, rationales, and test labels in Section 3 are produced by an LLM generating search queries and reasoning over results; no manual quality audit, inter-annotator agreement, or error analysis is reported.
  • domain assumption The LLM's claim decomposition and relevance filtering reliably isolate atomic claims from long-form responses.
    Section 3 follows Wei et al. (2024) to split long responses into atomized claims and filter relevant ones; errors in decomposition propagate to both training and evaluation.
  • domain assumption Balanced accuracy computed on a small held-out split is a meaningful comparison metric.
    The full set yields 2,711 claims split into 70%, 20%, and 10% partitions, so the test set is roughly 270 claims; BAcc point estimates without confidence intervals may be noisy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Auxiliary Tasks Improves Reference-Free Hallucination Detection in Open-Domain Long-Form Generation." pith.science (2026). https://pith.science/paper/4N6V5UML

@misc{pith2026250512265,
  author       = {Pith},
  title        = {Pith review of: Learning Auxiliary Tasks Improves Reference-Free Hallucination Detection in Open-Domain Long-Form Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4N6V5UML}},
  note         = {Machine review of arXiv:2505.12265}
}
read the original abstract

Hallucination, the generation of factually incorrect information, remains a significant challenge for large language models (LLMs), especially in open-domain long-form generation. Existing approaches for detecting hallucination in long-form tasks either focus on limited domains or rely heavily on external fact-checking tools, which may not always be available. In this work, we systematically investigate reference-free hallucination detection in open-domain long-form responses. Our findings reveal that internal states (e.g., model's output probability and entropy) alone are insufficient for reliably (i.e., better than random guessing) distinguishing between factual and hallucinated content. To enhance detection, we explore various existing approaches, including prompting-based methods, probing, and fine-tuning, with fine-tuning proving the most effective. To further improve the accuracy, we introduce a new paradigm, named RATE-FT, that augments fine-tuning with an auxiliary task for the model to jointly learn with the main task of hallucination detection. With extensive experiments and analysis using a variety of model families & datasets, we demonstrate the effectiveness and generalizability of our method, e.g., +3% over general fine-tuning methods on LongFact.

Figures

Figures reproduced from arXiv: 2505.12265 by the authors.

Figure 1
Figure 1. Comparison between Fine-Tuning and RATE [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Hallucination detection results based on token [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 5
Figure 5. Hallucination detection results based on the [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: Prompts for different prompting methods. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Prompts used for different components of [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Model’s Pfactual after applying RATE-FT for both factual and hallucinated claims. Xiaokang Zhang, Zijun Yao, Jing Zhang, Kaifeng Yun, Jifan Yu, Juanzi Li, and Jie Tang. 2024b. Transfer￾able and efficient non-factual content detection via probe training with offline con…
Figure 9
Figure 9. Figure 9: Prompt for extracting the original output given [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 6 canonical work pages

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Vaibhav Adlakha, Parishad BehnamGhader, Xing Han Lu, Nicholas Meade, and Siva Reddy. 2024. https://doi.org/10.1162/tacl_a_00667 Evaluating correctness and faithfulness of instruction-following models for question answering . Transactions of the Association for Computational Linguistics, 12:681--699

    doi:10.1162/tacl_a_00667 2024

  4. [4]

    David Paul Ausubel. 2012. https://books.google.com/books?hl=zh-CN&lr=&id=wfckBAAAQBAJ&oi=fnd&pg=PR9&dq=Consolidating+knowledge+through+repetition+in+diverse+contexts+is+a+fundamental+principle+of+effective+human+learning.&ots=mbZAesQ1GP&sig=K4asgsTJsZOrERmDQBfPw5lYREQ#v=onepage&q&f=false The acquisition and retention of knowledge: A cognitive view . Sprin...

    work page 2012

  5. [5]

    Amanda Bertsch, Alex Xie, Graham Neubig, and Matthew Gormley. 2023. https://doi.org/10.18653/v1/2023.bigpicture-1.9 It ' s MBR all the way down: Modern generation techniques through the lens of minimum B ayes risk . In Proceedings of the Big Picture Workshop, pages 108--122, Singapore. Association for Computational Linguistics

    doi:10.18653/v1/2023.bigpicture-1.9 2023

  6. [6]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert - Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litw...

    2020

  7. [7]

    Jifan Chen, Grace Kim, Aniruddh Sriram, Greg Durrett, and Eunsol Choi. 2024 a . https://doi.org/10.18653/v1/2024.naacl-long.196 Complex claim verification with evidence retrieved in the wild . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Pa...

    doi:10.18653/v1/2024.naacl-long.196 2024

  8. [8]

    Lida Chen, Zujie Liang, Xintao Wang, Jiaqing Liang, Yanghua Xiao, Feng Wei, Jinglei Chen, Zhenghong Hao, Bing Han, and Wei Wang. 2024 b . https://arxiv.org/abs/2406.10881 Teaching large language models to express knowledge boundary from their own signals . arXiv preprint arXiv:2406.10881

    arXiv 2024

Show all 48 references
  1. [9]

    Yijie Chen, Yijin Liu, Fandong Meng, Yufeng Chen, Jinan Xu, and Jie Zhou. 2023. https://arxiv.org/abs/2308.12674 Improving translation faithfulness of large language models via augmenting instructions . arXiv preprint arXiv:2308.12674

    2023 arXiv

  2. [10]

    Glass, and Pengcheng He

    Yung-Sung Chuang, Yujia Xie, Hongyin Luo, Yoon Kim, James R. Glass, and Pengcheng He. 2024. https://openreview.net/forum?id=Th6NyL07na Dola: Decoding by contrasting layers improves factuality in large language models . In The Twelfth International Conference on Learning Repres...

    2024

  3. [11]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. 2024. https://arxiv.org/abs/2407.21783 The llama 3 herd of models . arXiv preprint arXiv:2407.21783

    2024 arXiv

  4. [12]

    Ekaterina Fadeeva, Aleksandr Rubashevskii, Artem Shelmanov, Sergey Petrakov, Haonan Li, Hamdy Mubarak, Evgenii Tsymbalov, Gleb Kuzmin, Alexander Panchenko, Timothy Baldwin, Preslav Nakov, and Maxim Panov. 2024. https://doi.org/10.18653/v1/2024.findings-acl.558 Fact-checking th...

    2024 doi

  5. [13]

    Suriya Gunasekar, Yi Zhang, Jyoti Aneja, Caio Cesar Teodoro Mendes, Allie Del Giorno, Sivakanth Gopi, Mojan Javaheripi, Piero Conti Kauffmann, Gustavo Henrique de Rosa, Olli Saarikivi, Adil Salim, Shital Shah, Harkirat Behl, Xin Wang, Sebastien Bubeck, Ronen Eldan, Adam Tauman...

    2024

  6. [14]

    Danny Hernandez, Tom Brown, Tom Conerly, Nova DasSarma, Dawn Drain, Sheer El-Showk, Nelson Elhage, Zac Hatfield-Dodds, Tom Henighan, Tristan Hume, et al. 2022. https://arxiv.org/abs/2205.10487 Scaling laws and interpretability of learning from repeated data . arXiv preprint ar...

    2022 arXiv

  7. [15]

    Minda Hu, Bowei He, Yufei Wang, Liangyou Li, Chen Ma, and Irwin King. 2024. https://arxiv.org/abs/2406.11267 Mitigating large language model hallucination with faithful finetuning . arXiv preprint arXiv:2406.11267

    2024 arXiv

  8. [16]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, et al. 2024. https://arxiv.org/abs/2311.05232 A survey on hallucination in large language models: Principles, taxonomy, challenges, and open qu...

    2024 arXiv

  9. [17]

    Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. https://arxiv.org/abs/2310.06825 Mistral 7b . arXiv preprint arXiv:2310.06825

    2023 arXiv

  10. [18]

    Nikhil Kandpal, Haikang Deng, Adam Roberts, Eric Wallace, and Colin Raffel. 2023. https://proceedings.mlr.press/v202/kandpal23a.html Large language models struggle to learn long-tail knowledge . In Proceedings of the 40th International Conference on Machine Learning, volume 20...

    2023

  11. [19]

    Cheongwoong Kang and Jaesik Choi. 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.518 Impact of co-occurrence on factual knowledge of large language models . In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 7721--7735, Singapore. Associatio...

    2023 doi

  12. [20]

    Sanyam Kapoor, Nate Gruver, Manley Roberts, Katherine Collins, Arka Pal, Umang Bhatt, Adrian Weller, Samuel Dooley, Micah Goldblum, and Andrew Gordon Wilson. 2024. https://arxiv.org/abs/2406.08391 Large language models must be taught to know what they don't know . arXiv prepri...

    2024 arXiv

  13. [21]

    Philippe Laban, Wojciech Kry \'s ci \'n ski, Divyansh Agarwal, Alexander R Fabbri, Caiming Xiong, Shafiq Joty, and Chien-Sheng Wu. 2023. https://arxiv.org/abs/2305.14540 Llms as factual reasoners: Insights from existing benchmarks and beyond . arXiv preprint arXiv:2305.14540

    2023 arXiv

  14. [22]

    Bennett, and Marti A

    Philippe Laban, Tobias Schnabel, Paul N. Bennett, and Marti A. Hearst. 2022. https://doi.org/10.1162/tacl_a_00453 S umma C : Re-visiting NLI -based models for inconsistency detection in summarization . Transactions of the Association for Computational Linguistics, 10:163--177

    2022 doi

  15. [23]

    Faisal Ladhak, Esin Durmus, Mirac Suzgun, Tianyi Zhang, Dan Jurafsky, Kathleen McKeown, and Tatsunori Hashimoto. 2023. https://doi.org/10.18653/v1/2023.eacl-main.234 When do pre-training biases propagate to downstream tasks? a case study in text summarization . In Proceedings ...

    2023 doi

  16. [24]

    Daliang Li, Ankit Singh Rawat, Manzil Zaheer, Xin Wang, Michal Lukasik, Andreas Veit, Felix Yu, and Sanjiv Kumar. 2023 a . https://doi.org/10.18653/v1/2023.findings-acl.112 Large language models with controllable working memory . In Findings of the Association for Computationa...

    2023 doi

  17. [25]

    Kenneth Li, Oam Patel, Fernanda Vi \'e gas, Hanspeter Pfister, and Martin Wattenberg. 2023 b . https://openreview.net/forum?id=aLLuYpn83y Inference-time intervention: Eliciting truthful answers from a language model . In Thirty-seventh Conference on Neural Information Processi...

    2023

  18. [26]

    Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

    Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024. https://doi.org/10.1162/tacl_a_00638 Lost in the middle: How language models use long contexts . Transactions of the Association for Computational Linguistics, 12...

    2024 doi

  19. [27]

    Mat \'e o Mahaut, Laura Aina, Paula Czarnowska, Momchil Hardalov, Thomas M \"u ller, and Lluis Marquez. 2024. https://doi.org/10.18653/v1/2024.acl-long.250 Factual confidence of LLM s: on reliability and robustness of current estimators . In Proceedings of the 62nd Annual Meet...

    2024 doi

  20. [28]

    Potsawee Manakul, Adian Liusie, and Mark Gales. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.557 S elf C heck GPT : Zero-resource black-box hallucination detection for generative large language models . In Proceedings of the 2023 Conference on Empirical Methods in Natural...

    2023 doi

  21. [29]

    Kevin Meng, Arnab Sen Sharma, Alex J Andonian, Yonatan Belinkov, and David Bau. 2023. https://openreview.net/forum?id=MkbcAHIYgyS Mass-editing memory in a transformer . In The Eleventh International Conference on Learning Representations

    2023

  22. [30]

    Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen-tau Yih, Pang Koh, Mohit Iyyer, Luke Zettlemoyer, and Hannaneh Hajishirzi. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.741 FA ct S core: Fine-grained atomic evaluation of factual precision in long form text generatio...

    2023 doi

  23. [31]

    Feng Nan, Ramesh Nallapati, Zhiguo Wang, Cicero Nogueira dos Santos, Henghui Zhu, Dejiao Zhang, Kathleen McKeown, and Bing Xiang. 2021. https://doi.org/10.18653/v1/2021.eacl-main.235 Entity-level factual consistency of abstractive text summarization . In Proceedings of the 16t...

    2021 doi

  24. [32]

    Zhihong Shao, Yeyun Gong, Yelong Shen, Minlie Huang, Nan Duan, and Weizhu Chen. 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.620 Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy . In Findings of the Association for Comput...

    2023 doi

  25. [33]

    Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R. Bowman, Esin DURMUS, Zac Hatfield-Dodds, Scott R Johnston, Shauna M Kravec, Timothy Maxwell, Sam McCandlish, Kamal Ndousse, Oliver Rausch, Nicholas Schiefer, Da Yan, Miranda Zhang, and Ethan Pere...

    2024

  26. [34]

    Weijia Shi, Xiaochuang Han, Mike Lewis, Yulia Tsvetkov, Luke Zettlemoyer, and Wen-tau Yih. 2024 a . https://doi.org/10.18653/v1/2024.naacl-short.69 Trusting your evidence: Hallucinate less with context-aware decoding . In Proceedings of the 2024 Conference of the North America...

    2024 doi

  27. [35]

    Smith, Luke Zettlemoyer, Wen tau Yih, and Mike Lewis

    Weijia Shi, Sewon Min, Maria Lomeli, Chunting Zhou, Margaret Li, Xi Victoria Lin, Noah A. Smith, Luke Zettlemoyer, Wen tau Yih, and Mike Lewis. 2024 b . https://openreview.net/forum?id=LXVswInHOo In-context pretraining: Language modeling beyond document boundaries . In The Twe...

    2024

  28. [36]

    Weihang Su, Changyue Wang, Qingyao Ai, Yiran Hu, Zhijing Wu, Yujia Zhou, and Yiqun Liu. 2024. https://doi.org/10.18653/v1/2024.findings-acl.854 Unsupervised real-time hallucination detection based on the internal states of large language models . In Findings of the Association...

    2024 doi

  29. [37]

    Liyan Tang, Philippe Laban, and Greg Durrett. 2024. https://arxiv.org/abs/2404.10774 Minicheck: Efficient fact-checking of llms on grounding documents . arXiv preprint arXiv:2404.10774

    2024 arXiv

  30. [38]

    David Wan, Mengwen Liu, Kathleen McKeown, Markus Dreyer, and Mohit Bansal. 2023. https://doi.org/10.18653/v1/2023.eacl-main.210 Faithfulness-aware decoding strategies for abstractive summarization . In Proceedings of the 17th Conference of the European Chapter of the Associati...

    2023 doi

  31. [39]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. https://arxiv.org/abs/2201.11903 Chain-of-thought prompting elicits reasoning in large language models . Advances in neural information processing systems, 35:24824--24837

    2022 arXiv

  32. [40]

    Jerry Wei, Da Huang, Yifeng Lu, Denny Zhou, and Quoc V Le. 2023. https://arxiv.org/abs/2308.03958 Simple synthetic data reduces sycophancy in large language models . arXiv preprint arXiv:2308.03958

    2023 arXiv

  33. [41]

    Jerry Wei, Chengrun Yang, Xinying Song, Yifeng Lu, Nathan Hu, Dustin Tran, Daiyi Peng, Ruibo Liu, Da Huang, Cosmo Du, et al. 2024. https://arxiv.org/abs/2403.18802 Long-form factuality in large language models . arXiv preprint arXiv:2403.18802

    2024 arXiv

  34. [42]

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, et al. 2024. https://arxiv.org/abs/2407.10671 Qwen2 technical report . arXiv preprint arXiv:2407.10671

    2024 arXiv

  35. [43]

    Yakir Yehuda, Itzik Malkiel, Oren Barkan, Jonathan Weill, Royi Ronen, and Noam Koenigstein. 2024. https://doi.org/10.18653/v1/2024.acl-long.506 I nterrogate LLM : Zero-resource hallucination detection in LLM -generated answers . In Proceedings of the 62nd Annual Meeting of the...

    2024 doi

  36. [44]

    Muru Zhang, Ofir Press, William Merrill, Alisa Liu, and Noah A. Smith. 2024 a . https://openreview.net/forum?id=FPlaQyAGHu How language model hallucinations can snowball . In Forty-first International Conference on Machine Learning

    2024

  37. [45]

    Xiaokang Zhang, Zijun Yao, Jing Zhang, Kaifeng Yun, Jifan Yu, Juanzi Li, and Jie Tang. 2024 b . https://doi.org/10.18653/v1/2024.acl-long.668 Transferable and efficient non-factual content detection via probe training with offline consistency checking . In Proceedings of the 6...

    2024 doi

  38. [46]

    Yue Zhang, Yafu Li, Leyang Cui, Deng Cai, Lemao Liu, Tingchen Fu, Xinting Huang, Enbo Zhao, Yu Zhang, Yulong Chen, et al. 2023. https://arxiv.org/abs/2309.01219 Siren's song in the ai ocean: a survey on hallucination in large language models . arXiv preprint arXiv:2309.01219

    2023 arXiv

  39. [47]

    Yiran Zhao, Jinghan Zhang, I Chern, Siyang Gao, Pengfei Liu, Junxian He, et al. 2023. https://arxiv.org/abs/2310.00741 Felm: Benchmarking factuality evaluation of large language models . Advances in Neural Information Processing Systems, 36

    2023 arXiv

  40. [48]

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, and Zheyan Luo. 2024. https://doi.org/10.18653/v1/2024.acl-demos.38 L lama F actory: Unified efficient fine-tuning of 100+ language models . In Proceedings of the 62nd Annual Meeting of the Association for Computational Lin...

    2024 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.