Pith. sign in

REVIEW 3 major objections 6 minor 5 cited by

RbFT: Robust Fine-tuning for Retrieval-Augmented Generation against Retrieval Defects

T0 review · 3 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that fine-tuning an LLM on two tasks—classifying each retrieved document as useful, noisy, irrelevant, or counterfactual, and answering correctly from defective contexts—makes RAG far more resilient to retrieval defects…

desk verdict Useful two-task fine-tuning recipe for RAG robustness, but the headline gains are in-distribution; needs held-out defect tests before the generality claim holds. read the letter →

arxiv 2501.18365 v1 pith:42YVF27V submitted 2025-01-30 cs.CL cs.IR

Yiteng Tu , Weihang Su , Yujia Zhou , Yiqun Liu , Qingyao Ai This is my paper · ORCID
classification cs.CLcs.IR
keywords retrieval-augmentedgenerationrobustfine-tuningretrievaldefectsdefectdetectionutilityextractionlow-rankadaptationopen-domainquestionansweringcounterfactualdocuments
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

Retrieval-augmented generation inherits whatever defects the retriever and knowledge base have, and this paper argues that the root fix is to make the language model itself more skeptical and more resourceful. RbFT fine-tunes a 3B-parameter LLM on two tasks: Defects Detection, which labels each retrieved document as helpful, possibly relevant but unhelpful, irrelevant, or containing incorrect information, and Utility Extraction, which trains the model to produce the correct answer from defective retrieval lists. On Natural Questions, HotpotQA, and TriviaQA, LoRA-fine-tuned Llama-3.2-3B and Qwen2.5-3B models outperform prior robustness methods in every tested setting, with the largest margins when all retrieved documents are defective; exact-match gains over the second-best method reach 72.4% (Llama) and 64.1% (Qwen) on counterfactual defects. Because only weights change, per-query inference time stays essentially identical to vanilla RAG's, unlike multi-step robustness pipelines. If the result holds, retrieval-defect robustness can be bought with training data rather than with slower inference schemes.

What carries the argument

The engine of RbFT is a pair of fine-tuning tasks trained jointly with low-rank adaptation (LoRA), which updates a small set of parameters and keeps training efficient. Defects Detection uses a listwise prompt that forces the model to output one of four assessments for each document: it helps answer the question, it is possibly relevant but does not help, it is irrelevant, or it contains incorrect information. Utility Extraction keeps the standard answer-generation objective but feeds the model the defective document list, so the model learns to ignore or look past bad evidence and draw on the useful fraction or on its own parametric knowledge. Defective documents are injected at probability $\tau \in \{0.2, 0.4, 0.6, 0.8, 1.0\}$, with noisy documents sampled from below rank 50, irrelevant documents sampled from the full corpus, and counterfactual documents generated by having Llama-3.2-3B-Instruct produce a plausible wrong answer and then rewriting the passages around that wrong answer. What carries the argument is the combination of the two tasks: the detection labels tell the model which inputs to distrust, while the answer-generation task gives it practice succeeding despite them, and the ablation shows each task alone is weaker.

What would settle it

Take the same RbFT-fine-tuned models and evaluate them on defect types absent from training—character-level adversarial perturbations, documents poisoned by a different attack algorithm, or naturally imperfect retrievals from a different retriever and corpus—and compare exact-match accuracy against vanilla RAG. If the improvement over vanilla RAG shrinks to zero or reverses on these held-out defects, the central robustness claim is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that one fine-tuning recipe can make a RAG system substantially more robust to defective retrieved documents without sacrificing clean-set performance or inference speed. Concretely, RbFT uses the same corrupted document lists for both tasks: assess each document and then answer the question despite the bad context. The severity of defects is controlled by a replacement probability $\tau$ over the top-5 retrieved list; the Hard setting is $\tau = 1.0$, where every document is defective. Table 1 reports that under Hard counterfactual defects, RbFT reaches EM 33.8 (Llama) and 25.1 (Qwen) versus vanilla RAG's 9.0 and 8.9, and the relative improvement over the second-best method is 72.4% and 64.1%. The authors read the mechanism as teaching the model to judge input quality and to extract and synthesize only useful evidence; their attention case study shows smoother attention over documents after fine-tuning, with less focus on misleading tokens.

Load-bearing premise

The load-bearing premise is that the synthetic defects used for training and evaluation—lower-ranked passages, random corpus passages, and passages rewritten around an LLM-generated wrong answer—are representative of the retrieval defects a real RAG system will meet, so the model learns a general defensive skill rather than memorizing those specific corruption patterns.

Editorial extensions

If this is right

  • A 3B-parameter LLM can keep much of its accuracy even when the retriever returns zero useful documents; in the Hard+Mix condition RbFT scores EM 31.9 (Llama) and 24.0 (Qwen) while vanilla RAG scores 11.4 and 11.0.
  • Robustness does not have to cost latency: RbFT's inference time is 0.196 s per query for both models, essentially equal to vanilla RAG, compared with 1.207 s for RobustRAG and 3.417 s for AstuteRAG.
  • Because RbFT is a weight-only change, it can be stacked with inference-time defenses rather than replacing them, which the paper identifies as a route to further robustness.
  • A single model trained across multiple $\tau$ values handles both mild and extreme defect rates, so a deployed system need not know the defect severity in advance.

Reading between the lines

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

  • The evaluation uses only defects the authors synthesized, so whether RbFT teaches a general 'distrust and verify' policy or memorizes those corruption patterns is untested; a held-out evaluation with adversarial typos, poisoning attacks, or naturally occurring retrieval failures would resolve this.
  • The per-document assessments produced by Defects Detection could double as a free explanation or logging signal, or as training data for a separate reranker, but the paper does not propose or evaluate any such use.
  • The counterfactual generator is the same model family as one of the fine-tuned test models (Llama-3.2-3B-Instruct), so an open question is whether robustness persists against counterfactuals written by a different or stronger author; the paper does not test this.
  • The clean-setting gain over vanilla RAG suggests the retriever's top-5 already carries false positives, implying part of RbFT's effect is correcting ordinary retrieval noise rather than only extreme corruption; this could be checked by re-running the recipe with a stronger retriever and seeing whether the clean gain shrinks.
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 / 6 minor

Summary. The paper proposes RbFT, a LoRA-based fine-tuning method for RAG that trains a generation model on two auxiliary tasks: Defect Detection (classify each retrieved document as helpful, noisy, irrelevant, or counterfactual) and Utility Extraction (answer the query given defective documents). The method is evaluated on NQ, HotpotQA, and TriviaQA with three synthetic defect types injected under a replacement probability tau. Experiments report that RbFT outperforms Vanilla RAG and four inference-time robustness baselines (RobustRAG, CRAG, InstructRAG, AstuteRAG) in both EM and F1, and that it does so at essentially the same inference cost as Vanilla RAG.

Significance. If the reported results hold, the paper makes a useful practical contribution: a lightweight fine-tuning recipe that improves robustness to noisy, irrelevant, and counterfactual retrieved content while preserving in-distribution accuracy and introducing no extra inference overhead. The method is clearly specified, the code and data are promised to be released, and the in-distribution gains on the synthetic defect benchmark are large and consistent across two base models. The significance is conditional, however, on the synthetic defect process being representative: because the evaluation shares the training distribution of defect types and tau values, the current experiments do not establish the advertised generality to 'diverse retrieval conditions' or to real-world retrieval errors.

major comments (3)
  1. [§5.3, §5.4] The central robustness claim is not supported by the current evaluation protocol. The training and evaluation use the same synthetic defect-generation pipeline: noisy documents are sampled from lower-ranked results of the same e5-base-v2 retrieval list, irrelevant documents are sampled from the same corpus, counterfactual documents are generated by Llama-3.2-3B-Instruct, and Section 5.4 states that 'during the evaluation phase, the same tau values are used.' As a result, the improvements in Table 1 and Figure 4 measure performance on the exact defect distribution the model was optimized for, rather than robustness to a range of retrieval conditions. To substantiate the abstract's claim of 'diverse retrieval conditions,' the authors need to evaluate on held-out defect types (e.g., adversarial perturbations, human-written or independently generated misinformation), on a different retriever (e.g., BM25 or a different dense retriever), and at tau values not seen in training.
  2. [§5.2, Table 1] The comparison with baselines is asymmetric and conflates method effect with fine-tuning. RbFT is fine-tuned on the target datasets and the synthetic defect distribution, whereas RobustRAG, CRAG, InstructRAG, and AstuteRAG are applied without any fine-tuning on these datasets. The reported improvement ratios (for example, 72.4% EM over the second-best method in the Hard+Counterfactual setting for Llama) therefore may reflect the general benefit of task-specific adaptation rather than the specific Defect Detection and Utility Extraction objectives. The authors should include a control model fine-tuned with the same LoRA hyperparameters and the same data but with a standard QA objective (the Vanilla+UE ablation is a partial control but is not compared against the inference-time baselines in Table 1), as well as, ideally, an existing RAG fine-tuning baseline such as RA-DIT.
  3. [§5.3 (counterfactual generation)] For the Llama variant, counterfactual documents are generated by Llama-3.2-3B-Instruct, the same model that is subsequently fine-tuned. Since the Defect Detection task provides the model with labels for these exact generated rewrites, the model may learn to recognize the generator's stylistic or lexical fingerprints rather than a general ability to detect counterfactual content. The counterfactual results in Table 1 and Figure 4 should be re-verified with counterfactual documents generated by a different model (e.g., Qwen or GPT-4) or with human-written misinformation to exclude this shortcut.
minor comments (6)
  1. [Table 2] The change-ratio notation in Table 2 is ambiguous: for example, Vanilla+DD reaches 49.7 EM with an '↑2.7%' label in the Clean setting, which is actually better than RbFT's 48.4 EM, yet the text describes single-task variants as weaker. Please clarify the sign convention and, where a single-task variant outperforms RbFT, discuss the implication for the two-task design.
  2. [Figure 4] The six curves in each panel of Figure 4 are difficult to distinguish in grayscale; using distinct markers and line styles would make the figure readable in print.
  3. [§4, §5.4] The paper does not specify the exact prompting protocol used at evaluation time: it should state explicitly that the Utility Extraction prompt alone is used for generation and that the Defect Detection labels are used only as a training signal, or describe how the two heads are combined at inference.
  4. [§5.4] The description of fine-tuning omits the data mixing ratio between the Defect Detection and Utility Extraction tasks, the maximum sequence length, and the number of training steps; adding these would make the training procedure fully reproducible from the text alone.
  5. [§6.4] The phrase 'RbFT is vertical to these methods' is nonstandard; it should be 'orthogonal to' or 'complementary to,' and the claim of compatibility would be stronger if backed by an explicit integration experiment or by a reference to existing evidence.
  6. [Figure 2 caption] There is a typo in the Figure 2 caption: 'Emperical' should be 'Empirical.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: RbFT's gains are empirical and measured on held-out queries; the matched train/eval defect distribution is a generalization caveat, not a circular reduction.

full rationale

RbFT's central claim is empirical: fine-tuning with Defect Detection and Utility Extraction improves exact-match scores under corrupted retrieval inputs. The training pipeline (Sections 4.1–4.2) optimizes the standard next-token objective (Eq. 3) over training queries, and the LoRA weights are not fitted to the 3,000 evaluation queries or to the reported EM values. Evaluation in Section 6 uses test or validation splits (1,000 queries per dataset) that are disjoint from the 20,000 training instances, and the robustness gain is measured rather than defined by the training objective. The closest candidate for circularity is the matched defect-generation process: Section 5.4 states that during evaluation the same tau values are used, and both training and evaluation draw noisy, irrelevant, and counterfactual documents from the same Section 5.3 pipeline. That matching limits external generalization because the model may be learning the synthetic corruption distribution, but it is not a logical reduction: the tau values are experimental conditions, not fitted parameters, and the test answers are never used during training. No load-bearing self-citation or imported uniqueness claim was found; same-group references in Section 2 are related-work context, not premises of the derivation. Therefore no circular step is exhibited.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central claim rests on several domain assumptions about the representativeness of synthetic defects and the transferability of fine-tuning, but it introduces no invented entities and only standard supervised-learning assumptions. The free parameters are simulation and hyperparameter choices, not fitted constants in a derivation.

free parameters (5)
  • defect replacement probability tau = {0.2, 0.4, 0.6, 0.8, 1.0} in training and evaluation
    Chosen by the authors to simulate defect severity; the same values are used in training and test, making the evaluation in-distribution.
  • noisy document sampling rank threshold = documents ranked after 50 are treated as noisy
    Hand-chosen proxy for hard negatives, without validation against real noisy retrievals.
  • irrelevant document sampling strategy = random documents from the full corpus
    Hand-chosen proxy for random negatives; no evidence this matches real irrelevant retrievals.
  • counterfactual generation model = Llama-3.2-3B-Instruct
    The same model family as one of the base LLMs is used to create misleading documents, which may bias the trained model's familiarity with the corruption pattern.
  • LoRA hyperparameters = rank=16, alpha=64, lr=1e-5, batch=16, 2 epochs
    Reported without a tuning study; their effect on the result is not analyzed.
assumptions (3)
  • domain assumption Synthetic defect distribution (noisy, irrelevant, counterfactual and their mixture) is representative of real-world retrieval defects.
    The paper constructs defects via negative sampling and LLM rewriting and does not validate against naturally occurring defective retrievals.
  • domain assumption Fine-tuning on NQ, HQA, and TQA transfers to other RAG applications.
    Only three QA datasets are evaluated, all with the same retriever and Wikipedia corpus.
  • domain assumption The LLM's improved defect detection and utility extraction learned through the two prompts generalize beyond the exact prompt formats.
    Inference uses a QA prompt; the model is never evaluated on whether it can explicitly flag defects at test time.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RbFT: Robust Fine-tuning for Retrieval-Augmented Generation against Retrieval Defects." pith.science (2026). https://pith.science/paper/42YVF27V

@misc{pith2026250118365,
  author       = {Pith},
  title        = {Pith review of: RbFT: Robust Fine-tuning for Retrieval-Augmented Generation against Retrieval Defects},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/42YVF27V}},
  note         = {Machine review of arXiv:2501.18365}
}
read the original abstract

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by integrating external knowledge retrieved from a knowledge base. However, its effectiveness is fundamentally constrained by the reliability of both the retriever and the knowledge base. In real-world scenarios, imperfections in these components often lead to the retrieval of noisy, irrelevant, or misleading counterfactual information, ultimately undermining the trustworthiness of RAG systems. To address this challenge, we propose Robust Fine-Tuning (RbFT), a method designed to enhance the resilience of LLMs against retrieval defects through two targeted fine-tuning tasks. Experimental results demonstrate that RbFT significantly improves the robustness of RAG systems across diverse retrieval conditions, surpassing existing methods while maintaining high inference efficiency and compatibility with other robustness techniques.

Figures

Figures reproduced from arXiv: 2501.18365 by the authors.

Figure 1
Figure 1. Overview of our RbFT. Specifically, RbFT consists of two sub-tasks: Defects Detection and Utility Extraction, which [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Emperical study: the impact of different types of [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The effectiveness-robustness trade-off scatter dia [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The EM performance of all methods under 4 types of defective data with [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Case studies on the attention distribution over input documents of Vanilla RAG and RbFT under different retrieval [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 5 Pith papers

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

  1. Certified Domain Consistency for Multi-Domain Retrieval: Label-Free Per-Domain Contamination Control with Conformal Risk Guarantees

    cs.LG 2026-07 conditional novelty 6.0 of 10

    C3R certifies per-domain retrieval contamination budgets using a two-split conformal scheme, without query-time domain labels.

  2. Magic Mushroom: A Customizable Benchmark for Fine-grained Analysis of Retrieval Noise Erosion in RAG Systems

    cs.CL 2025-06 conditional novelty 6.0 of 10

    A configurable benchmark with four retrieval-noise types shows RAG accuracy drops sharply beyond 50% noise and that noise type, not just quantity, determines failure patterns.

  3. Decoupling Reasoning and Knowledge Injection for In-Context Knowledge Editing

    cs.CL 2025-05 conditional novelty 6.0 of 10

    DecKER decouples reasoning from knowledge editing by planning with masked placeholders before retrieving edited facts, improving multi-hop QA accuracy after knowledge edits.

  4. Curriculum Guided Reinforcement Learning for Efficient Multi Hop Retrieval Augmented Generation

    cs.CL 2025-05 reject novelty 5.0 of 10

    EVO-RAG applies curriculum-guided reinforcement learning with time-varying reward weights to multi-hop RAG, reporting improved EM on HotpotQA, 2WikiMultiHopQA, and MuSiQue.

  5. Dynamic and Parametric Retrieval-Augmented Generation

    cs.CL 2025-06 unverdicted novelty 2.0 of 10

    A tutorial outline that categorizes recent RAG work into Dynamic RAG and Parametric RAG, and explains why both are needed.

Reference graph

Works this paper leans on

62 extracted references · 10 canonical work pages · cited by 5 Pith papers

  1. [1]

    Md Adnan Arefeen, Biplob Debnath, and Srimat Chakradhar. 2024. Leancontext: Cost-efficient domain-specific question answering using llms. Natural Language Processing Journal 7 (2024), 100065

    work page 2024

  2. [2]

    Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Ruther- ford, Katie Millican, George Bm Van Den Driessche, Jean-Baptiste Lespiau, Bog- dan Damoc, Aidan Clark, et al. 2022. Improving language models by retrieving from trillions of tokens. In International conference on machine learning . PMLR, 2206–2240

    2022

  3. [3]

    Harsh Chaudhari, Giorgio Severi, John Abascal, Matthew Jagielski, Christopher A Choquette-Choo, Milad Nasr, Cristina Nita-Rotaru, and Alina Oprea. 2024. Phan- tom: General Trigger Attacks on Retrieval Augmented Language Generation. arXiv preprint arXiv:2405.20485 (2024)

    arXiv 2024

  4. [4]

    Jiawei Chen, Hongyu Lin, Xianpei Han, and Le Sun. 2024. Benchmarking large language models in retrieval-augmented generation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 17754–17762

    2024

  5. [5]

    Wenhu Chen, Hexiang Hu, Xi Chen, Pat Verga, and William W Cohen. 2022. Murag: Multimodal retrieval-augmented generator for open question answering over images and text. arXiv preprint arXiv:2210.02928 (2022)

    arXiv 2022

  6. [6]

    Sukmin Cho, Soyeong Jeong, Jeongyeon Seo, Taeho Hwang, and Jong C Park

  7. [7]

    Sunhao Dai, Chen Xu, Shicheng Xu, Liang Pang, Zhenhua Dong, and Jun Xu. 2024. Unifying Bias and Unfairness in Information Retrieval: A Survey of Challenges and Opportunities with Large Language Models. arXiv preprint arXiv:2404.11457 (2024)

    arXiv 2024

  8. [8]

    Zhuyun Dai, Vincent Y Zhao, Ji Ma, Yi Luan, Jianmo Ni, Jing Lu, Anton Bakalov, Kelvin Guu, Keith B Hall, and Ming-Wei Chang. 2022. Promptagator: Few-shot dense retrieval from 8 examples. arXiv preprint arXiv:2209.11755 (2022)

    arXiv 2022

Show all 62 references
  1. [9]

    Qian Dong, Qingyao Ai, Hongning Wang, Yiding Liu, Haitao Li, Weihang Su, Yiqun Liu, Tat-Seng Chua, and Shaoping Ma. 2025. Decoupling Knowledge and Context: An Efficient and Effective Retrieval Augmented Generation Framework via Cross Attention. In Proceedings of the ACM on Web...

    2025

  2. [10]

    Yibing Du, Antoine Bosselut, and Christopher D Manning. 2022. Synthetic disinformation attacks on automated fact verification systems. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 36. 10581–10589

    2022

  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. The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    2024 arXiv

  4. [12]

    Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, and Jonathan Larson. 2024. From local to global: A graph rag approach to query-focused summarization. arXiv preprint arXiv:2404.16130 (2024)

    2024 arXiv

  5. [13]

    Michael Glass, Gaetano Rossiello, Md Faisal Mahbub Chowdhury, Ankita Rajaram Naik, Pengshan Cai, and Alfio Gliozzo. 2022. Re2G: Retrieve, rerank, generate. arXiv preprint arXiv:2207.06300 (2022)

    2022 arXiv

  6. [14]

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang. 2020. Retrieval augmented language model pre-training. In International conference on machine learning. PMLR, 3929–3938

    2020

  7. [15]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 (2021)

    2021 arXiv

  8. [16]

    Yuntong Hu, Zhihan Lei, Zheng Zhang, Bo Pan, Chen Ling, and Liang Zhao. 2024. GRAG: Graph Retrieval-Augmented Generation. arXiv preprint arXiv:2405.16506 (2024)

    2024 arXiv

  9. [17]

    Siqing Huo, Negar Arabzadeh, and Charles Clarke. 2023. Retrieving support- ing evidence for generative question answering. In Proceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region . 11–20

    2023

  10. [18]

    Gautier Izacard and Edouard Grave. 2020. Leveraging passage retrieval with generative models for open domain question answering. arXiv preprint arXiv:2007.01282 (2020)

    2020 arXiv

  11. [19]

    Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. 2023. Atlas: Few-shot learning with retrieval augmented language models. Journal of Machine Learning Research 24, 251 ...

    2023

  12. [20]

    Zhengbao Jiang, Luyu Gao, Jun Araki, Haibo Ding, Zhiruo Wang, Jamie Callan, and Graham Neubig. 2022. Retrieval as attention: End-to-end learning of retrieval and reading within a single transformer. arXiv preprint arXiv:2212.02027 (2022)

    2022 arXiv

  13. [21]

    Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. 2017. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension. arXiv preprint arXiv:1705.03551 (2017)

    2017 arXiv

  14. [22]

    Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open- domain question answering. arXiv preprint arXiv:2004.04906 (2020)

    2020 arXiv

  15. [23]

    Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, et al. 2019. Natural questions: a benchmark for question answering research. Transactions of the Association for C...

    2019

  16. [24]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing S...

    2020

  17. [25]

    Huayang Li, Yixuan Su, Deng Cai, Yan Wang, and Lemao Liu. 2022. A survey on retrieval-augmented text generation. arXiv preprint arXiv:2202.01110 (2022)

    2022 arXiv

  18. [26]

    Weitao Li, Junkai Li, Weizhi Ma, and Yang Liu. 2024. Citation-Enhanced Genera- tion for LLM-based Chatbot. arXiv preprint arXiv:2402.16063 (2024)

    2024 arXiv

  19. [27]

    Xinbei Ma, Yeyun Gong, Pengcheng He, Hai Zhao, and Nan Duan. 2023. Query rewriting for retrieval-augmented large language models. arXiv preprint arXiv:2305.14283 (2023)

    2023 arXiv

  20. [28]

    Yixiao Ma, Yueyue Wu, Weihang Su, Qingyao Ai, and Yiqun Liu. 2023. CaseEn- coder: A Knowledge-enhanced Pre-trained Model for Legal Case Encoding.arXiv preprint arXiv:2305.05393 (2023)

    2023 arXiv

  21. [29]

    Rui Mao, Guanyi Chen, Xulang Zhang, Frank Guerin, and Erik Cambria. 2023. GPTEval: A survey on assessments of ChatGPT and GPT-4. arXiv preprint arXiv:2308.12488 (2023)

    2023 arXiv

  22. [30]

    Liangming Pan, Wenhu Chen, Min-Yen Kan, and William Yang Wang. 2021. Attacking open-domain question answering by injecting misinformation. arXiv preprint arXiv:2110.07803 (2021)

    2021 arXiv

  23. [31]

    Yikang Pan, Liangming Pan, Wenhu Chen, Preslav Nakov, Min-Yen Kan, and William Yang Wang. 2023. On the risk of misinformation pollution with large language models. arXiv preprint arXiv:2305.13661 (2023)

    2023 arXiv

  24. [32]

    Boci Peng, Yun Zhu, Yongchao Liu, Xiaohe Bo, Haizhou Shi, Chuntao Hong, Yan Zhang, and Siliang Tang. 2024. Graph retrieval-augmented generation: A survey. arXiv preprint arXiv:2408.08921 (2024)

    2024 arXiv

  25. [33]

    Yingqi Qu, Yuchen Ding, Jing Liu, Kai Liu, Ruiyang Ren, Wayne Xin Zhao, Daxi- ang Dong, Hua Wu, and Haifeng Wang. 2020. RocketQA: An optimized training approach to dense passage retrieval for open-domain question answering. arXiv preprint arXiv:2010.08191 (2020). RbFT: Robust ...

    2020 arXiv

  26. [34]

    Stephen Robertson, Hugo Zaragoza, et al . 2009. The probabilistic relevance framework: BM25 and beyond. Foundations and Trends® in Information Retrieval 3, 4 (2009), 333–389

    2009

  27. [35]

    Alireza Salemi and Hamed Zamani. 2024. Towards a search engine for machines: Unified ranking for multiple retrieval-augmented large language models. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval . 741–751

    2024

  28. [36]

    Rulin Shao, Jacqueline He, Akari Asai, Weijia Shi, Tim Dettmers, Sewon Min, Luke Zettlemoyer, and Pang Wei Koh. 2024. Scaling Retrieval-Based Language Models with a Trillion-Token Datastore. arXiv preprint arXiv:2407.12854 (2024)

    2024 arXiv

  29. [37]

    Weihang Su, Qingyao Ai, Xiangsheng Li, Jia Chen, Yiqun Liu, Xiaolong Wu, and Shengluan Hou. 2023. Wikiformer: Pre-training with Structured Information of Wikipedia for Ad-hoc Retrieval. arXiv preprint arXiv:2312.10661 (2023)

    2023 arXiv

  30. [38]

    Weihang Su, Qingyao Ai, Yueyue Wu, Yixiao Ma, Haitao Li, and Yiqun Liu. 2023. Caseformer: Pre-training for Legal Case Retrieval.arXiv preprint arXiv:2311.00333 (2023)

    2023 arXiv

  31. [39]

    Weihang Su, Yiran Hu, Anzhe Xie, Qingyao Ai, Quezi Bing, Ning Zheng, Yun Liu, Weixing Shen, and Yiqun Liu. 2024. STARD: A Chinese Statute Retrieval Dataset Derived from Real-life Queries by Non-professionals. InFindings of the Association for Computational Linguistics: EMNLP 2...

    2024 doi

  32. [40]

    Weihang Su, Xiangsheng Li, Yiqun Liu, Min Zhang, and Shaoping Ma. 2023. Thuir2 at ntcir-16 session search (ss) task. arXiv preprint arXiv:2307.00250 (2023)

    2023 arXiv

  33. [41]

    Weihang Su, Yichen Tang, Qingyao Ai, Changyue Wang, Zhijing Wu, and Yiqun Liu. 2024. Mitigating entity-level hallucination in large language models. In Proceedings of the 2024 Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the...

    2024

  34. [42]

    Weihang Su, Yichen Tang, Qingyao Ai, Zhijing Wu, and Yiqun Liu. 2024. Dragin: Dynamic retrieval augmented generation based on the real-time information needs of large language models. arXiv preprint arXiv:2403.10081 (2024)

    2024 arXiv

  35. [43]

    Weihang Su, Yichen Tang, Qingyao Ai, Junxi Yan, Changyue Wang, Hongning Wang, Ziyi Ye, Yujia Zhou, and Yiqun Liu. 2025. Parametric Retrieval Augmented Generation. arXiv preprint arXiv:2501.15915 (2025)

    2025 arXiv

  36. [44]

    Weihang Su, Changyue Wang, Qingyao Ai, Yiran Hu, Zhijing Wu, Yujia Zhou, and Yiqun Liu. 2024. Unsupervised real-time hallucination detection based on the internal states of large language models. arXiv preprint arXiv:2403.06448 (2024)

    2024 arXiv

  37. [45]

    Chao-Hong Tan, Jia-Chen Gu, Chongyang Tao, Zhen-Hua Ling, Can Xu, Huang Hu, Xiubo Geng, and Daxin Jiang. 2022. TegTok: Augmenting text generation via task-specific and open-world knowledge. arXiv preprint arXiv:2203.08517 (2022)

    2022 arXiv

  38. [46]

    Changyue Wang, Weihang Su, Qingyao Ai, and Yiqun Liu. 2024. Knowledge Editing through Chain-of-Thought. arXiv preprint arXiv:2412.17727 (2024)

    2024 arXiv

  39. [47]

    Changyue Wang, Weihang Su, Hu Yiran, Qingyao Ai, Yueyue Wu, Cheng Luo, Yiqun Liu, Min Zhang, and Shaoping Ma. 2024. LeKUBE: A Legal Knowledge Update BEnchmark. arXiv preprint arXiv:2407.14192 (2024)

    2024 arXiv

  40. [48]

    Fei Wang, Xingchen Wan, Ruoxi Sun, Jiefeng Chen, and Sercan Ö Arık. 2024. As- tute rag: Overcoming imperfect retrieval augmentation and knowledge conflicts for large language models. arXiv preprint arXiv:2410.07176 (2024)

    2024 arXiv

  41. [49]

    Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. 2022. Text embeddings by weakly-supervised contrastive pre-training. arXiv preprint arXiv:2212.03533 (2022)

    2022 arXiv

  42. [50]

    Shuo Wang, Shotaro Kinoshita, and Hiromi M Yokoyama. 2022. Performance of ChatGPT and GPT-4 on neurosurgery written board examinations. Neurosurgery (2022), 10–1227

    2022

  43. [51]

    Zhepei Wei, Wei-Lin Chen, and Yu Meng. 2024. InstructRAG: Instructing Re- trieval Augmented Generation via Self-Synthesized Rationales. arXiv preprint arXiv:2406.13629 (2024)

    2024 arXiv

  44. [52]

    Orion Weller, Aleem Khan, Nathaniel Weir, Dawn Lawrie, and Benjamin Van Durme. 2022. Defending Against Misinformation Attacks in Open-Domain Question Answering. arXiv preprint arXiv:2212.10002 (2022)

    2022 arXiv

  45. [53]

    Chong Xiang, Tong Wu, Zexuan Zhong, David Wagner, Danqi Chen, and Prateek Mittal. 2024. Certifiably Robust RAG against Retrieval Corruption. arXiv preprint arXiv:2405.15556 (2024)

    2024

  46. [54]

    Shi-Qi Yan, Jia-Chen Gu, Yun Zhu, and Zhen-Hua Ling. 2024. Corrective retrieval augmented generation. arXiv preprint arXiv:2401.15884 (2024)

    2024 arXiv

  47. [55]

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al . 2024. Qwen2. 5 Technical Report. arXiv preprint arXiv:2412.15115 (2024)

    2024 arXiv

  48. [56]

    Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W Cohen, Ruslan Salakhutdinov, and Christopher D Manning. 2018. HotpotQA: A dataset for di- verse, explainable multi-hop question answering. arXiv preprint arXiv:1809.09600 (2018)

    2018 arXiv

  49. [57]

    ChengXiang Zhai. 2008. Statistical language models for information retrieval. Synthesis lectures on human language technologies 1, 1 (2008), 1–141

    2008

  50. [58]

    Jingtao Zhan, Jiaxin Mao, Yiqun Liu, Jiafeng Guo, Min Zhang, and Shaoping Ma. 2021. Optimizing dense retrieval model training with hard negatives. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval . 1503–1512

    2021

  51. [59]

    Yue Zhang, Yafu Li, Leyang Cui, Deng Cai, Lemao Liu, Tingchen Fu, Xinting Huang, Enbo Zhao, Yu Zhang, Yulong Chen, et al . 2023. Siren’s song in the AI ocean: a survey on hallucination in large language models. arXiv preprint arXiv:2309.01219 (2023)

    2023 arXiv

  52. [60]

    Yujia Zhou, Yan Liu, Xiaoxi Li, Jiajie Jin, Hongjin Qian, Zheng Liu, Chaozhuo Li, Zhicheng Dou, Tsung-Yi Ho, and Philip S Yu. 2024. Trustworthiness in retrieval- augmented generation systems: A survey. arXiv preprint arXiv:2409.10102 (2024)

    2024 arXiv

  53. [61]

    Wei Zou, Runpeng Geng, Binghui Wang, and Jinyuan Jia. 2024. Poisonedrag: Knowledge poisoning attacks to retrieval-augmented generation of large lan- guage models. arXiv preprint arXiv:2402.07867 (2024)

    2024 arXiv

  54. [2024]

    arXiv preprint arXiv:2404.13948 (2024)

    Typos that Broke the RAG’s Back: Genetic Attack on RAG Pipeline by Simulating Documents in the Wild via Low-level Perturbations. arXiv preprint arXiv:2404.13948 (2024)

    2024 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.