CQC-RAG: Robust Retrieval-Augmented Generation via Cross-Query Consistency

Jie Shao; Sifan Liu; Yanjia Sun

arxiv: 2606.13438 · v1 · pith:O6KZJEP5new · submitted 2026-06-11 · 💻 cs.IR

CQC-RAG: Robust Retrieval-Augmented Generation via Cross-Query Consistency

Yanjia Sun , Sifan Liu , Jie Shao This is my paper

Pith reviewed 2026-06-27 05:34 UTC · model grok-4.3

classification 💻 cs.IR

keywords Retrieval-Augmented GenerationCross-Query ConsistencyHallucination DetectionMulti-Query ReasoningQuestion AnsweringRobustnessEvidence Grounding

0 comments

The pith

CQC-RAG filters hallucinations in retrieval-augmented generation by checking whether candidate answers keep stable high confidence across syntactically varied but meaning-equivalent queries.

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

The paper starts from the observation that standard RAG is brittle because one query can retrieve misleading evidence and because multi-query methods often rely on random decoding or single-view scoring. It proposes that correct answers will show high and stable confidence when the same fact is asked in several different but equivalent ways, while hallucinated answers will fluctuate. The resulting CQC-RAG system rewrites the input into diverse queries, builds query-specific contexts from a shared document pool, extracts evidence-grounded answer pairs, and retains only the answers whose confidence remains consistent. On four open-domain QA benchmarks this produces gains of 4.76 EM points on TriviaQA and 9.12 EM points on MuSiQue over the strongest prior multi-query baseline.

Core claim

The Cross-Query Consistency Hypothesis states that correct answers maintain high confidence across semantically equivalent but syntactically diverse queries, whereas noise-induced hallucinations exhibit unstable confidence under such query variations. CQC-RAG operationalizes the hypothesis by rewriting the original question, reranking a shared document pool to form query-conditioned contexts, applying an evidence-grounded extraction protocol, and selecting the answer with the most stable confidence across those contexts, thereby performing self-evaluation without external supervision or expanded retrieval.

What carries the argument

The Cross-Query Consistency Hypothesis together with the co-designed pipeline of query rewriting, shared-pool reranking, evidence-grounded pair extraction, and stability-based selection.

If this is right

Multi-query RAG no longer needs external voting or expanded retrieval sets; internal consistency across query variants suffices for selection.
The same consistency signal can be used to reject answers that are unstable even when they appear confident on the original query.
The method works on existing LLMs without additional training or supervision.
Gains appear on both single-hop (TriviaQA) and multi-hop (MuSiQue) datasets, suggesting the mechanism is not limited to one reasoning depth.

Where Pith is reading between the lines

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

If the hypothesis holds, the same stability check could be applied to chain-of-thought traces or tool-use sequences to detect internal contradictions.
The approach might be combined with retrieval re-ranking that explicitly favors documents supporting multiple query variants.
A practical deployment could cache the rewritten queries and their contexts so that repeated similar questions incur little extra cost.

Load-bearing premise

That an answer which is factually correct will reliably produce higher and more stable model confidence when the question is rephrased in different syntactic forms.

What would settle it

A controlled test in which known correct answers are shown to drop in confidence or change identity when the input question is rewritten into semantically equivalent variants while keeping the retrieved evidence fixed.

Figures

Figures reproduced from arXiv: 2606.13438 by Jie Shao, Sifan Liu, Yanjia Sun.

**Figure 1.** Figure 1: Illustration of CQC-RAG. CQC-RAG employs a parallel rewriting mechanism to construct multiple queries that are [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Figure 2: Performance comparison on the TriviaQA and [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

read the original abstract

Retrieval-Augmented Generation (RAG) has become a common approach for improving the factuality of Large Language Models (LLMs), yet its reliability remains highly sensitive to how external evidence is retrieved and used. Semantically equivalent queries with different syntactic forms may lead to different retrieval results, while irrelevant or misleading documents can further induce hallucinated answers. Existing multi-path reasoning methods improve robustness by sampling multiple candidate answers and applying voting- or confidence-based selection, but they still face two limitations: diversity is often injected through uncontrollable decoding randomness, and answer evaluation is usually confined to a single query-induced evidence view. To address these limitations, we propose a Cross-Query Consistency Hypothesis: correct answers tend to maintain high confidence across semantically equivalent but syntactically diverse queries, whereas noise-induced hallucinations exhibit unstable confidence under such query variations. Based on this hypothesis, we introduce CQC-RAG, a framework that co-designs query-level diversity injection with cross-query consistency evaluation. CQC-RAG rewrites the original question into diverse but meaning-preserving queries, reranks a shared document pool to construct query-conditioned reasoning contexts, applies an evidence-grounded protocol to extract answer-evidence pairs and selects answers according to their confidence stability across these contexts. This design enables self-evaluation without external supervision and does not rely on expanded retrieval coverage. Experiments on four open-domain question answering benchmarks show that CQC-RAG outperforms the strongest previous multi-query baseline by +4.76 pp EM on TriviaQA and +9.12 pp EM on MuSiQue, validating the effectiveness of cross-query consistency for filtering noise-induced hallucinations.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

CQC-RAG gets measurable gains on TriviaQA and MuSiQue from query rewriting plus stability selection, but the abstract gives no direct evidence that consistency filtering is what drives the improvement.

read the letter

The paper's main move is the Cross-Query Consistency Hypothesis plus the CQC-RAG pipeline that rewrites queries, reranks a shared document pool, and picks answers by confidence stability across those rewrites. This is a bit more structured than plain multi-path sampling or voting, and the shared pool keeps retrieval cost from growing. The reported numbers are concrete: +4.76 EM on TriviaQA and +9.12 on MuSiQue over the strongest prior multi-query baseline.

The work is straightforward and addresses a real deployment pain point without extra supervision or retrieval budget. The design choices around shared reranking and evidence-grounded answer extraction are practical.

The soft spot is exactly the one in the stress-test note. The abstract claims the hypothesis is validated by the end-to-end gains, yet supplies no stability distributions split by correctness, no variance statistics, and no ablation that isolates the consistency step from the rewriting and reranking. Without those, the delta could come from the diversity injection alone. The full paper may contain the missing checks, but nothing in the visible material shows them.

This is for people already working on RAG robustness or multi-query methods. A reader in that area would get value from the pipeline description and the benchmark deltas. The paper is coherent on its own terms and cites the relevant prior work.

It deserves peer review so the mechanism can be examined directly.

Referee Report

1 major / 1 minor

Summary. The paper proposes the Cross-Query Consistency Hypothesis—that correct answers maintain stable high confidence across syntactically diverse but semantically equivalent queries, while hallucinations induced by noise exhibit unstable confidence—and introduces the CQC-RAG framework that implements this via query rewriting for diversity, shared-pool reranking, evidence-grounded answer extraction, and selection by cross-query confidence stability. It reports empirical gains of +4.76 pp EM on TriviaQA and +9.12 pp EM on MuSiQue over the strongest prior multi-query baseline, claiming validation of the hypothesis on four open-domain QA benchmarks without external supervision or expanded retrieval.

Significance. If the reported gains are attributable to the consistency-based filtering mechanism rather than query diversity alone, the work would offer a practical self-evaluation approach for improving RAG factuality. The co-design of diversity injection and stability evaluation addresses a recognized limitation in existing multi-path methods, but the absence of direct hypothesis tests limits the assessed impact to incremental empirical improvement pending further validation.

major comments (1)

[Abstract] Abstract: The central claim that the results validate the Cross-Query Consistency Hypothesis rests on aggregate EM improvements (+4.76 pp on TriviaQA, +9.12 pp on MuSiQue) versus a multi-query baseline. No direct measurements (e.g., confidence stability distributions or variance statistics conditioned on answer correctness) or ablations that isolate the consistency filter from query rewriting and reranking are described, leaving open the possibility that gains arise from increased query diversity alone.

minor comments (1)

[Abstract] Abstract: The text states experiments on four benchmarks but reports detailed gains only for TriviaQA and MuSiQue; the results for the remaining two benchmarks should be summarized for completeness.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comment below.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim that the results validate the Cross-Query Consistency Hypothesis rests on aggregate EM improvements (+4.76 pp on TriviaQA, +9.12 pp on MuSiQue) versus a multi-query baseline. No direct measurements (e.g., confidence stability distributions or variance statistics conditioned on answer correctness) or ablations that isolate the consistency filter from query rewriting and reranking are described, leaving open the possibility that gains arise from increased query diversity alone.

Authors: We appreciate the referee's observation. The strongest prior multi-query baseline already incorporates query diversity through multiple paths, so the reported gains over this baseline are attributable to the addition of cross-query consistency filtering rather than diversity alone. That said, we agree that direct measurements (such as confidence stability distributions and variance statistics conditioned on answer correctness) and an ablation isolating the consistency filter would provide stronger, more explicit support for the hypothesis. In the revised version we will add these analyses, including the requested statistics and ablation, and will adjust the abstract to reflect the expanded evidence. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical hypothesis and benchmark evaluation remain independent

full rationale

The paper states the Cross-Query Consistency Hypothesis explicitly as a new premise, then describes a co-designed framework (query rewriting, reranking, consistency-based selection) whose effectiveness is measured by end-to-end EM gains on TriviaQA and MuSiQue. No equations, fitted parameters, or self-referential definitions appear that would make any reported quantity equivalent to its inputs by construction. No self-citations are invoked as load-bearing uniqueness theorems. The derivation chain is therefore self-contained against external benchmarks and does not reduce to tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the validity of the stated hypothesis and the assumption that query rewriting produces sufficiently diverse yet meaning-preserving variants whose confidence patterns reliably separate signal from noise; no free parameters or invented entities are named in the abstract.

axioms (1)

domain assumption Correct answers tend to maintain high confidence across semantically equivalent but syntactically diverse queries, whereas noise-induced hallucinations exhibit unstable confidence under such query variations.
This is explicitly introduced in the abstract as the Cross-Query Consistency Hypothesis that motivates the entire framework.

pith-pipeline@v0.9.1-grok · 5820 in / 1336 out tokens · 25484 ms · 2026-06-27T05:34:20.237167+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

34 extracted references · 11 canonical work pages · 4 internal anchors

[1]

Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024

2024
[2]

Tianyu Cao, Neel Bhandari, Akhila Yerukola, Akari Asai, and Maarten Sap. 2026. Out of Style: RAG’s Fragility to Linguistic Variation. InProceedings of the 19th CQC-RAG: Robust Retrieval-Augmented Generation via Cross-Query Consistency Conference’17, July 2017, Washington, DC, USA Conference of the European Chapter of the Association for Computational Lin-...

2026
[3]

Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu
[4]

BGE M3-Embedding: Multi-Lingual, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation.CoRRabs/2402.03216 (2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024
[5]

Bernal Jimenez Gutierrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su
[6]

HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models. InAdvances in Neural Information Processing Systems 38: An- nual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024

2024
[7]

Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave
[8]

Atlas: Few-shot Learning with Retrieval Augmented Language Models.J. Mach. Learn. Res.24 (2023), 251:1–251:43

2023
[9]

Soyeong Jeong, Jinheon Baek, Sukmin Cho, Sung Ju Hwang, and Jong Park
[10]

Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity. InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), NAACL 2024, Mexico City, Mexico, June 16-21, 2024. 7036–7050

2024
[11]

Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, De- vendra Singh Chaplot, Diego de Las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, and William El Sayed. 2023. Mistral 7B.CoRRabs/2310.06...

work page internal anchor Pith review Pith/arXiv arXiv 2023
[12]

Weld, and Luke Zettlemoyer

Mandar Joshi, Eunsol Choi, Daniel S. Weld, and Luke Zettlemoyer. 2017. TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehen- sion. InProceedings of the 55th Annual Meeting of the Association for Computational Linguistics, ACL 2017, Vancouver, Canada, July 30 - August 4, Volume 1: Long Papers. 1601–1611

2017
[13]

Zhewei Kang, Xuandong Zhao, and Dawn Song. 2025. Scalable Best-of-N Se- lection for Large Language Models via Self-Certainty.CoRRabs/2502.18581 (2025)

work page arXiv 2025
[14]

Seongyun Lee, Yongrae Jo, Minju Seo, Moontae Lee, and Minjoon Seo. 2026. Lost in the Noise: How Reasoning Models Fail with Contextual Distractors.CoRR abs/2601.07226 (2026)

work page arXiv 2026
[15]

Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. InAdvances in Neural Information Processing Systems 33: Annual Conference on Neural Information...

2020
[16]

Zhicong Li, Jiahao Wang, Zhishu Jiang, Hangyu Mao, Zhongxia Chen, Jiazhen Du, Yuanxing Zhang, Fuzheng Zhang, Di Zhang, and Yong Liu. 2024. DMQR-RAG: Diverse Multi-Query Rewriting for RAG.CoRRabs/2411.13154 (2024)

work page arXiv 2024
[17]

Xinbei Ma, Yeyun Gong, Pengcheng He, Hai Zhao, and Nan Duan. 2023. Query Rewriting for Retrieval-Augmented Large Language Models.CoRRabs/2305.14283 (2023)

work page arXiv 2023
[18]

Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memories. InProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2023, Toronto, Canada, July 9-1...

2023
[19]

Amir Taubenfeld, Tom Sheffer, Eran Ofek, Amir Feder, Ariel Goldstein, Zorik Gekhman, and Gal Yona. 2025. Confidence Improves Self-Consistency in LLMs. InFindings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025. 20090–20111

2025
[20]

Llama Team. 2024. The Llama 3 Herd of Models.CoRRabs/2407.21783 (2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024
[21]

Hieu Tran, Zonghai Yao, Zhichao Yang, Junda Wang, Yifan Zhang, Shuo Han, Feiyun Ouyang, and Hong Yu. 2025. RARE: Retrieval-Augmented Reasoning Enhancement for Large Language Models. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2025, Vienna, Austria, July 27 - August 1, 2025. 18305–18330

2025
[22]

Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal
[23]

MuSiQue: Multihop Questions via Single-hop Question Composition.Trans. Assoc. Comput. Linguistics10 (2022), 539–554

2022
[24]

Weiqin Wang, Yile Wang, and Hui Huang. 2025. Ranked Voting based Self- Consistency of Large Language Models. InFindings of the Association for Compu- tational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025. 14410– 14426

2025
[25]

Le, Ed H

Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-Consistency Improves Chain of Thought Reasoning in Language Models. InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023

2023
[26]

Le, Huaixiu Steven Zheng, Swaroop Mishra, Vincent Perot, Yuwei Zhang, Anush Mattapalli, Ankur Taly, Jingbo Shang, Chen- Yu Lee, and Tomas Pfister

Zilong Wang, Zifeng Wang, Long T. Le, Huaixiu Steven Zheng, Swaroop Mishra, Vincent Perot, Yuwei Zhang, Anush Mattapalli, Ankur Taly, Jingbo Shang, Chen- Yu Lee, and Tomas Pfister. 2025. Speculative RAG: Enhancing Retrieval Aug- mented Generation through Drafting. InThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore,...

2025
[27]

Siye Wu, Jian Xie, Jiangjie Chen, Tinghui Zhu, Kai Zhang, and Yanghua Xiao
[28]

How Easily do Irrelevant Inputs Skew the Responses of Large Language Models?CoRRabs/2404.03302 (2024)

work page arXiv 2024
[29]

An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jian Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keqin Bao, Kexin Yang, Le Yu, Liangha...

work page internal anchor Pith review Pith/arXiv arXiv 2025
[30]

Xizhong Yang, Haotian Zhang, Huiming Wang, and Mofei Song. 2026. Believe Your Model: Distribution-Guided Confidence Calibration.CoRRabs/2603.03872 (2026)

work page arXiv 2026
[31]

Yuxin Yang, Gangda Deng, Ömer Faruk Akgül, Nima Chitsazan, Yash Govilkar, Akasha Tigalappanavara, Shi-Xiong Zhang, Sambit Sahu, and Viktor Prasanna
[32]

SPARC-RAG: Adaptive Sequential-Parallel Scaling with Context Manage- ment for Retrieval-Augmented Generation.CoRRabs/2602.00083 (2026)

work page arXiv 2026
[33]

Fabbri, Gabriel Bernadett- Shapiro, Rui Zhang, Prasenjit Mitra, Caiming Xiong, and Chien-Sheng Wu

Nan Zhang, Prafulla Kumar Choubey, Alexander R. Fabbri, Gabriel Bernadett- Shapiro, Rui Zhang, Prasenjit Mitra, Caiming Xiong, and Chien-Sheng Wu. 2025. SiReRAG: Indexing Similar and Related Information for Multihop Reasoning. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025

2025
[34]

Xing, Hao Zhang, Joseph E

Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric P. Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. 2023. Judging LLM-as-a-Judge with MT- Bench and Chatbot Arena. InAdvances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems...

2023

[1] [1]

Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024

2024

[2] [2]

Tianyu Cao, Neel Bhandari, Akhila Yerukola, Akari Asai, and Maarten Sap. 2026. Out of Style: RAG’s Fragility to Linguistic Variation. InProceedings of the 19th CQC-RAG: Robust Retrieval-Augmented Generation via Cross-Query Consistency Conference’17, July 2017, Washington, DC, USA Conference of the European Chapter of the Association for Computational Lin-...

2026

[3] [3]

Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu

[4] [4]

BGE M3-Embedding: Multi-Lingual, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation.CoRRabs/2402.03216 (2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024

[5] [5]

Bernal Jimenez Gutierrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su

[6] [6]

HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models. InAdvances in Neural Information Processing Systems 38: An- nual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024

2024

[7] [7]

Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave

[8] [8]

Atlas: Few-shot Learning with Retrieval Augmented Language Models.J. Mach. Learn. Res.24 (2023), 251:1–251:43

2023

[9] [9]

Soyeong Jeong, Jinheon Baek, Sukmin Cho, Sung Ju Hwang, and Jong Park

[10] [10]

Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity. InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), NAACL 2024, Mexico City, Mexico, June 16-21, 2024. 7036–7050

2024

[11] [11]

Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, De- vendra Singh Chaplot, Diego de Las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, and William El Sayed. 2023. Mistral 7B.CoRRabs/2310.06...

work page internal anchor Pith review Pith/arXiv arXiv 2023

[12] [12]

Weld, and Luke Zettlemoyer

Mandar Joshi, Eunsol Choi, Daniel S. Weld, and Luke Zettlemoyer. 2017. TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehen- sion. InProceedings of the 55th Annual Meeting of the Association for Computational Linguistics, ACL 2017, Vancouver, Canada, July 30 - August 4, Volume 1: Long Papers. 1601–1611

2017

[13] [13]

Zhewei Kang, Xuandong Zhao, and Dawn Song. 2025. Scalable Best-of-N Se- lection for Large Language Models via Self-Certainty.CoRRabs/2502.18581 (2025)

work page arXiv 2025

[14] [14]

Seongyun Lee, Yongrae Jo, Minju Seo, Moontae Lee, and Minjoon Seo. 2026. Lost in the Noise: How Reasoning Models Fail with Contextual Distractors.CoRR abs/2601.07226 (2026)

work page arXiv 2026

[15] [15]

Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. InAdvances in Neural Information Processing Systems 33: Annual Conference on Neural Information...

2020

[16] [16]

Zhicong Li, Jiahao Wang, Zhishu Jiang, Hangyu Mao, Zhongxia Chen, Jiazhen Du, Yuanxing Zhang, Fuzheng Zhang, Di Zhang, and Yong Liu. 2024. DMQR-RAG: Diverse Multi-Query Rewriting for RAG.CoRRabs/2411.13154 (2024)

work page arXiv 2024

[17] [17]

Xinbei Ma, Yeyun Gong, Pengcheng He, Hai Zhao, and Nan Duan. 2023. Query Rewriting for Retrieval-Augmented Large Language Models.CoRRabs/2305.14283 (2023)

work page arXiv 2023

[18] [18]

Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memories. InProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2023, Toronto, Canada, July 9-1...

2023

[19] [19]

Amir Taubenfeld, Tom Sheffer, Eran Ofek, Amir Feder, Ariel Goldstein, Zorik Gekhman, and Gal Yona. 2025. Confidence Improves Self-Consistency in LLMs. InFindings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025. 20090–20111

2025

[20] [20]

Llama Team. 2024. The Llama 3 Herd of Models.CoRRabs/2407.21783 (2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024

[21] [21]

Hieu Tran, Zonghai Yao, Zhichao Yang, Junda Wang, Yifan Zhang, Shuo Han, Feiyun Ouyang, and Hong Yu. 2025. RARE: Retrieval-Augmented Reasoning Enhancement for Large Language Models. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2025, Vienna, Austria, July 27 - August 1, 2025. 18305–18330

2025

[22] [22]

Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal

[23] [23]

MuSiQue: Multihop Questions via Single-hop Question Composition.Trans. Assoc. Comput. Linguistics10 (2022), 539–554

2022

[24] [24]

Weiqin Wang, Yile Wang, and Hui Huang. 2025. Ranked Voting based Self- Consistency of Large Language Models. InFindings of the Association for Compu- tational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025. 14410– 14426

2025

[25] [25]

Le, Ed H

Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-Consistency Improves Chain of Thought Reasoning in Language Models. InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023

2023

[26] [26]

Le, Huaixiu Steven Zheng, Swaroop Mishra, Vincent Perot, Yuwei Zhang, Anush Mattapalli, Ankur Taly, Jingbo Shang, Chen- Yu Lee, and Tomas Pfister

Zilong Wang, Zifeng Wang, Long T. Le, Huaixiu Steven Zheng, Swaroop Mishra, Vincent Perot, Yuwei Zhang, Anush Mattapalli, Ankur Taly, Jingbo Shang, Chen- Yu Lee, and Tomas Pfister. 2025. Speculative RAG: Enhancing Retrieval Aug- mented Generation through Drafting. InThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore,...

2025

[27] [27]

Siye Wu, Jian Xie, Jiangjie Chen, Tinghui Zhu, Kai Zhang, and Yanghua Xiao

[28] [28]

How Easily do Irrelevant Inputs Skew the Responses of Large Language Models?CoRRabs/2404.03302 (2024)

work page arXiv 2024

[29] [29]

An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jian Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keqin Bao, Kexin Yang, Le Yu, Liangha...

work page internal anchor Pith review Pith/arXiv arXiv 2025

[30] [30]

Xizhong Yang, Haotian Zhang, Huiming Wang, and Mofei Song. 2026. Believe Your Model: Distribution-Guided Confidence Calibration.CoRRabs/2603.03872 (2026)

work page arXiv 2026

[31] [31]

Yuxin Yang, Gangda Deng, Ömer Faruk Akgül, Nima Chitsazan, Yash Govilkar, Akasha Tigalappanavara, Shi-Xiong Zhang, Sambit Sahu, and Viktor Prasanna

[32] [32]

SPARC-RAG: Adaptive Sequential-Parallel Scaling with Context Manage- ment for Retrieval-Augmented Generation.CoRRabs/2602.00083 (2026)

work page arXiv 2026

[33] [33]

Fabbri, Gabriel Bernadett- Shapiro, Rui Zhang, Prasenjit Mitra, Caiming Xiong, and Chien-Sheng Wu

Nan Zhang, Prafulla Kumar Choubey, Alexander R. Fabbri, Gabriel Bernadett- Shapiro, Rui Zhang, Prasenjit Mitra, Caiming Xiong, and Chien-Sheng Wu. 2025. SiReRAG: Indexing Similar and Related Information for Multihop Reasoning. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025

2025

[34] [34]

Xing, Hao Zhang, Joseph E

Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric P. Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. 2023. Judging LLM-as-a-Judge with MT- Bench and Chatbot Arena. InAdvances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems...

2023