Information Retrieval · arXiv:2609.19656 v1 · submitted 17 Sep 2026 · 55 references
Self-Evolving Search Index
The paper says
“Information retrieval is increasingly important as LLM agents tackle complex tasks involving diverse information needs.”
more of the abstract
Because retrieval relies on an index that represents each document through index keys, retrieval quality depends heavily on how effectively these keys expose the knowledge contained in each document. However, effective index representations vary across retrieval environments, making it difficult for any fixed optimization strategy to perform consistently. Yet evolving an index to its retrieval environment remains largely human-driven, requiring humans to diagnose retrieval failures, refine the optimization strategy, and reprocess the index accordingly. We propose SELF-INDEX, a framework that enables an index to self-evolve without human intervention. Its Optimizer autonomously diagnoses retrieval shortfalls, selectively revises the responsible index keys, and validates each revision before updating the index. Beyond reacting to observed retrieval demands, SELF-INDEX proactively explores additional demands through a Query Simulator, allowing the index to evolve beyond the queries already available for optimization. Across diverse corpora and retrievers, SELF-INDEX consistently improves retrieval performance while outperforming existing index optimization methods. We further show that these benefits extend to downstream applications, improving the effectiveness and efficiency of search agents and helping agent memory systems retrieve useful past interactions.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 18 Sep 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
The passes disagreed. Reader: holds up. Skeptic: agrees with the reader. Referee: major revision.
SELF-INDEX delivers an automated index-evolution loop with impressively broad experiments, but the LLM self-validation circularity is a real blind spot that needs human calibration. This line is wrong
3 major objections · 5 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 55 references say.
The paper says, at §A.1.4, Self-Validation; Table 5
The review did not quote this passage.
Pith says: Objection 1 of 3 · major
The faithfulness judge is the same LLM (Qwen3.6-35B-A3B) that performs diagnosis and revision, creating a potential circularity: the model may accept keys that are plausible but factually unsupported. The paper does not report any human evaluation of sampled keys to calibrate the judge's accuracy.
more of objection 1
The ablation (Table 5) only measures aggregate effect of removing the criterion, not false-positive or false-negative rates. A systematic failure mode (e.g., subtle factual distortion) could go undetected. The authors should either (a) provide a human evaluation on a representative sample, reporting acceptance rate and error types, or (b) explicitly discuss this limitation and provide evidence that the LLM judge is sufficiently reliable for the intended use.
The paper says, at §4.2, Table 1; §C.2, Table 14
The review did not quote this passage.
Pith says: Objection 2 of 3 · major
The paper claims SELF-INDEX 'consistently achieves the highest average score for every corpus type under every retriever.' While the average is highest across the three runs, the gain varies substantially across datasets (e.g., Biology gains >20 points while TheoremQA-Question gains <5 points). The authors should discuss this variability and its implications for the generalization claim. The current presentation with domain averages masks the distribution of improvements.
The paper says, at §3.3, Self-Exploration; §5, Figure 5
The review did not quote this passage.
Pith says: Objection 3 of 3 · major
The query simulator generates queries from the corpus using the same LLM backbone. While the paper shows that replacing the simulator with ReasonIR queries still yields gains (Figure 5), it does not rule out that the simulator's queries are biased toward the capabilities of that backbone.
more of objection 3
A stress test with a properly held-out corpus and a different query-generation method (e.g., crowdsourced) would strengthen the claim of generalizability. This is not a load-bearing flaw but limits the evidentiary weight for proactive evolution.
5 minor objections
The paper says, at Abstract
Pith says: minor objection 1 of 5
The abstract ends with '[CODE]' in the provided full text, which appears to be a remnant from compilation; please remove it.
The paper says, at Table 1 caption
Pith says: minor objection 2 of 5
The word 'Boldmarks' should be 'Bold marks' or 'Bold indicates'.
show 3 more
The paper says, at Figure 6 caption
Pith says: minor objection 3 of 5
The y-axis label reads 'nDCG@10' (missing a 'g'); should be 'nDCG@10'.
The paper says, at Appendix A.1.2
Pith says: minor objection 4 of 5
In Algorithm 2, the function BUILDPROFILES is not defined in the main text; a reference to the appendix is needed earlier.
The paper says, at Appendix C.2
Pith says: minor objection 5 of 5
Table 14 gives nDCG@10 averaged over three retrievers for each backbone; but the text describes average nDCG@10 and the numbers seem to be on different scales. Please clarify the aggregation formula.
What would settle it
Sample 200 accepted revision decisions from the optimized index (e.g., from BRIGHT Biology and Robotics). For each, have three human annotators assess: (a) whether the diagnosed shortfall is valid, (b) whether the proposed keys are supported by the document (faithfulness), (c) whether the keys are appropriately specific and separated. Compare human judgments to the system’s diagnosis, revision acceptance, and faithfulness score. Report the proportion of revisions that pass the system’s validation but fail human evaluation. If this proportion exceeds 10%, the self-validation loop is insufficiently reliable.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 18 Sep 2026
- Reader holds up
- Skeptic agrees
- Editor wrote the verdict above
- Referee major revision: 3 major objections, 5 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
SELF-INDEX proposes a clean loop: retrieve co-occurrence patterns, let an LLM diagnose and rewrite keys, then validate via faithfulness/specificity/separation checks. The experiments are impressively broad—multiple corpora (NL, code, math, tables), three retrievers, ablations, downstream agent tasks—and the gains are consistent and often large. The paper does a solid job of showing the method works across diverse settings, and the case studies provide helpful intuition.
The main soft spot is exactly the one in the stress test: the same LLM backbone (Qwen3.6-35B-A3B) handles diagnosis, revision, and faithfulness validation. The faithfulness judge scores the model's own output on a 0–3 scale. The paper does not include any human evaluation of sampled accepted revisions, so we don't know how often the loop admits keys that are plausible but factually unsupported. The ablation shows that removing criteria hurts performance in aggregate, but that doesn't measure per-key accuracy. That's a real limitation. To the authors' credit, separation and specificity checks are retrieval-based and independent of the LLM, which helps constrain the problem. The backbone sensitivity table (Table 14) shows that smaller models degrade but the gap narrows with size, suggesting the process isn't pure noise.
Overall, this is a substantive contribution to automated index optimization. The core idea is novel and the empirical support is strong. The self-validation circularity is a risk, but not a fatal flaw—the paper could address it with human sampling. I would send this to peer review and expect reviewers to ask for that calibration or a failure-mode analysis. It's worth citing for the architecture and breadth of results.
The strongest objection
The framework relies on the Qwen3.6-35B-A3B backbone for three critical tasks: diagnosing retrieval shortfalls, generating revised keys, and validating faithfulness. The faithfulness judge uses the same backbone to score each proposed key on a 0–3 scale, with a threshold of 2 for acceptance. This creates a potential circularity: if the model produces keys based on an incorrect understanding of the document, it may also judge those keys as faithful. The specificity and separation checks are retrieval-based and do not involve the LLM, but they cannot detect keys that are plausible but factually unsupported. The paper does not report a human evaluation of a sample of generated keys to verify that the validation criteria correctly accept only high-quality revisions. Without such calibration, it is unknown how often flawed keys pass validation and silently degrade retrieval in deployment scenarios beyond the tested distributions. The ablation study (Table 5) shows that removing any criterion reduces performance, but this does not measure the criterion’s actual accuracy—only its aggregate effect. A systematic failure mode (e.g., subtle factual distortion) could go undetected.
The reader's weakest assumption already identified the LLM backbone reliability as the core risk. The concern is indeed load-bearing: if the self-diagnosis and self-validation loop systematically produces low-quality keys that pass all three criteria, then the claimed gains may not generalize to new corpora or retrieval contexts. However, the paper provides substantial mitigating evidence: (1) consistent improvement across three retrievers, four corpus types, and multiple downstream tasks, (2) backbone sensitivity analysis showing that only models with ≥9B parameters work, suggesting capability threshold, (3) ablations showing each validation component contributes positively, (4) small standard errors across runs, and (5) the validation criteria include two retrieval-based checks (specificity and separation) that do not rely on LLM judgment. The missing piece is direct human verification of the revision quality. This gap does not invalidate the paper’s central claim—the empirical results are strong and diverse—but it does limit confidence about whether the framework is truly autonomous and reliable across unanticipated retrieval demands. Hence, the verdict should remain ACCEPT (unchanged) because the existing evidence is sufficient for publication, but the paper would be strengthened by adding a human evaluation of sampled revisions.
The referee's report
If the results are robust, SELF-INDEX offers a practical advancement by automating the labor-intensive process of index optimization without human supervision. The extensive experimental validation—spanning diverse corpora, retrievers, and downstream tasks—strengthens the claim of broad applicability. The modular design (optimizer + query simulator) is a novel contribution to the self-evolving paradigm for information retrieval.
Pith's reading of the paper (a machine's summary, not the authors')
This paper proposes SELF-INDEX, a framework that enables a retrieval index to evolve autonomously. It works by diagnosing retrieval failures from the index's own behaviour, selectively revising the index keys responsible, and validating each change before confirming it. The index also proactively explores new information needs through a query simulator, going beyond the queries it has already seen. Across natural language, code, math, and table corpora, and with three different retrievers, SELF-INDEX consistently improves retrieval quality and outperforms existing methods like Doc2Query, SPIKE, and RL-Index, often by large margins (up to 57% relative nDCG@10 gain). These gains carry over to downstream tasks: search agents answer more accurately with fewer calls, and agent memory systems retrieve more useful past interactions.
What carries the argument. The Optimizer's three-stage loop (Self-Diagnosis using co-retrieval profiles, Self-Revision targeting only diagnosed key sets, Self-Validation with three criteria) and the Query Simulator's Self-Exploration (document-grounded query generation with dissimilarity filtering). Together they allow the index to react to retrieval failures and proactively seek unrepresented needs.
Core claim. SELF-INDEX is a framework that makes a retrieval index self-evolve without human oversight. Its Optimizer runs a three-stage loop — Self-Diagnosis (identifies shortfalls from co-retrieval patterns), Self-Revision (rewrites only the diagnosed keys), and Self-Validation (checks faithfulness, specificity, and separation) — and its Query Simulator generates new queries to drive evolution beyond the observed ones. The central claim is that this autonomous, selective, and iterative process consistently produces better index keys than any fixed optimization strategy, across diverse corpora and retrievers, and that better keys improve downstream agent performance and memory utilization.
The premise it rests on. The LLM backbone (Qwen3.6-35B-A3B) used for diagnosis, revision, validation, and query simulation must generate reliable outputs without human oversight; if it hallucinates flawed diagnoses or invalid keys, the index evolution will fail.
If the paper is right (Pith's inferences, not the authors')
Index optimization can be fully automated, removing the human bottleneck of diagnosing failures and redesigning strategies.
A single index-evolution process can adapt to diverse corpora (natural language, code, math, tables) and retrievers (BM25, BGE-large, Qwen3-Emb) without manual reconfiguration.
Improved retrieval directly boosts the effectiveness and cost-efficiency of LLM search agents, as shown on BrowseComp-Plus.
Agent memory systems that rely on retrieval can reuse past experiences more effectively when their memory indices are evolved with SELF-INDEX.
The approach scales to larger corpora (tested up to 400K documents) while maintaining accuracy and reducing online costs.
The method could make high-quality retrieval accessible to non-experts who lack the resources to hand-craft index strategies.
The reliance on a strong LLM backbone (Qwen3.6-35B-A3B) for diagnosis and revision means performance may degrade on less capable models or on corpora with highly domain-specific language where the backbone's world knowledge is weak.
The Query Simulator's exploration might miss some rare but important queries; combining it with user-click feedback could broaden coverage further.
The self-evolution loop could be extended to other infrastructure components beyond search, such as database query optimizers or recommendation systems.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
Information retrieval is increasingly important as LLM agents tackle complex tasks involving diverse information needs. Because retrieval relies on an index that represents each document through index keys, retrieval quality depends heavily on how effectively these keys expose the knowledge contained in each document. However, effective index representations vary across retrieval environments, making it difficult for any fixed optimization strategy to perform consistently. Yet evolving an index to its retrieval environment remains largely human-driven, requiring humans to diagnose retrieval failures, refine the optimization strategy, and reprocess the index accordingly. We propose SELF-INDEX, a framework that enables an index to self-evolve without human intervention. Its Optimizer autonomously diagnoses retrieval shortfalls, selectively revises the responsible index keys, and validates each revision before updating the index. Beyond reacting to observed retrieval demands, SELF-INDEX proactively explores additional demands through a Query Simulator, allowing the index to evolve beyond the queries already available for optimization. Across diverse corpora and retrievers, SELF-INDEX consistently improves retrieval performance while outperforming existing index optimization methods. We further show that these benefits extend to downstream applications, improving the effectiveness and efficiency of search agents and helping agent memory systems retrieve useful past interactions.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-09-23
The authors' main claim, as the classifier picked it out 1.00
“Across diverse corpora and retrievers, SELF-INDEX consistently improves retrieval performance while outperforming existing index optimization methods.”
How the main claim is stated: asserted 1.00
What the abstract itself states, as probabilities: claims first 0.11 · states limitation 0.09 · human participants 0.06 · controlled comparison 0.60 · randomized assignment 0.02 · states quantitative result 0.04 · compares with prior results 0.94 · mentions code or data availability 0.03
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| Information retrieval is increasingly important as LLM agents tackle complex tasks involving diverse information needs. | background 1.00 | 0.04 | 0.02 | asserted 1.00 |
| Because retrieval relies on an index that represents each document through index keys, retrieval quality depends heavily on how effectively these keys expose the knowledge contained in each document. | background 1.00 | 0.06 | 0.02 | asserted 1.00 |
| However, effective index representations vary across retrieval environments, making it difficult for any fixed optimization strategy to perform consistently. | background 0.79 | 0.09 | 0.02 | asserted 0.99 |
| Yet evolving an index to its retrieval environment remains largely human-driven, requiring humans to diagnose retrieval failures, refine the optimization strategy, and reprocess the index accordingly. | gap 0.92 | 0.06 | 0.02 | asserted 1.00 |
| We propose SELF-INDEX, a framework that enables an index to self-evolve without human intervention. | contribution 1.00 | 0.84 | 0.19 | asserted 1.00 |
| Its Optimizer autonomously diagnoses retrieval shortfalls, selectively revises the responsible index keys, and validates each revision before updating the index. | method 0.97 | 0.76 | 0.08 | asserted 1.00 |
| Beyond reacting to observed retrieval demands, SELF-INDEX proactively explores additional demands through a Query Simulator, allowing the index to evolve beyond the queries already available for optimization. | method 0.88 | 0.86 | 0.05 | asserted 0.98 |
| Across diverse corpora and retrievers, SELF-INDEX consistently improves retrieval performance while outperforming existing index optimization methods. | result 1.00 | 0.95 | 0.66 | asserted 1.00 |
| We further show that these benefits extend to downstream applications, improving the effectiveness and efficiency of search agents and helping agent memory systems retrieve useful past interactions. | result 0.97 | 0.94 | 0.18 | asserted 1.00 |
What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.96
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 6,816 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (6)
Figures are the authors', reproduced from the paper.
References (55)
- [1] Imagine All The Relevance: Scenario-Profiled Indexing with Knowledge Expansion for Dense Retrieval , author=. Second Conference on Language Modeling , year= on Pith
- [2] Peter Baile Chen and Tomer Wolfson and Mike Cafarella and Dan Roth , booktitle=. EnrichIndex: Using. 2025 , url= on Pith
- [3] RL-Index: Reinforcement Learning for Retrieval Index Reasoning , author=. 2026 , eprint= on Pith
- [4] Document Expansion by Query Prediction , author=. 2019 , eprint= on Pith
- [5] Chen, Yanfei and Yoon, Jinsung and Sachan, Devendra Singh and Wang, Qingze and Cohen-Addad, Vincent and Bateni, Mohammadhossein and Lee, Chen-Yu and Pfister, Tomas. Re-Invoke: Tool Invocation Rewriting for Zero-Shot Tool Retrieval. Findings of the Association for Computational... on Pithdoi:10.18653/v1/2024.findings-emnlp.270
- [6] Jeong, Soyeong and Baek, Jinheon and Park, ChaeHun and Park, Jong. Unsupervised Document Expansion for Information Retrieval with Stochastic Text Generation. Proceedings of the Second Workshop on Scholarly Document Processing. 2021. doi:10.18653/v1/2021.sdp-1.2 on Pithdoi:10.18653/v1/2021.sdp-1.2
- [7] Boudin, Florian and Gallina, Ygor and Aizawa, Akiko. Keyphrase Generation for Scientific Document Retrieval. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2020. doi:10.18653/v1/2020.acl-main.105 on Pithdoi:10.18653/v1/2020.acl-main.105
- [8] Chen, Tong and Wang, Hongwei and Chen, Sihao and Yu, Wenhao and Ma, Kaixin and Zhao, Xinran and Zhang, Hongming and Yu, Dong. Dense X Retrieval: What Retrieval Granularity Should We Use?. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2... on Pithdoi:10.18653/v1/2024.emnlp-main.845
- [9] AutoIndex: Learning Representation Programs for Retrieval , author=. 2026 , eprint= on Pith
- [10] Huang, Chengsong and Yu, Wenhao and Wang, Xiaoyang and Zhang, Hongming and Li, Zongxia and Li, Ruosen and Huang, Jiaxin and Mi, Haitao and Yu, Dong , booktitle =. R-Zero: Self-Evolving Reasoning LLM from Zero Data , url = on Pith
- [11] Dr. Zero: Self-Evolving Search Agents without Training Data , author=. 2026 , eprint= on Pith
- [12] SPICE: Self-Play In Corpus Environments Improves Reasoning , author=. 2025 , eprint= on Pith
- [13] Zhao, Andrew and Wu, Yiran and Wu, Tong and Xu, Quentin and Yue, Yang and Lin, Matthieu and Wang, Shenzhi and Wu, Qingyun and Zheng, Zilong and Huang, Gao , booktitle =. Absolute Zero: Reinforced Self-play Reasoning with Zero Data , url =. doi:10.52202/085713-3534 , editor = on Pithdoi:10.52202/085713-3534
- [14] Tool-R0: Self-Evolving LLM Agents for Tool-Learning from Zero Data , author=. 2026 , eprint= on Pith
- [15] Raifer, Nimrod and Raiber, Fiana and Tennenholtz, Moshe and Kurland, Oren , title =. Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2017 , isbn =. doi:10.1145/3077136.3080785 , abstract = on PitharXiv
- [16] Gerald, Thomas and Soulier, Laure , title =. Advances in Information Retrieval: 44th European Conference on IR Research, ECIR 2022, Stavanger, Norway, April 10–14, 2022, Proceedings, Part I , pages =. 2022 , isbn =. doi:10.1007/978-3-030-99736-6_17 , abstract = on Pithdoi:10.1007/978-3-030-99736-6_17
- [17] Son, HuiJeong and Kang, Hyeongu and Kim, Sunho and Ho, Subeen and Kang, SeongKu and Lee, Dongha and Yoon, Susik , title =. Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 , pages =. 2026 , isbn =. doi:10.1145/3770854.3780281 , abstract = on PitharXiv
- [18] Xiao, Shitao and Liu, Zheng and Zhang, Peitian and Muennighoff, Niklas and Lian, Defu and Nie, Jian-Yun , title =. Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2024 , isbn =. doi:10.1145/3626772.3657... on PitharXiv
- [19] Robertson, Stephen and Zaragoza, Hugo , title =. Found. Trends Inf. Retr. , month = apr, pages =. 2009 , issue_date =. doi:10.1561/1500000019 , abstract = on Pithdoi:10.1561/1500000019
- [20] Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models , author=. arXiv preprint arXiv:2506.05176 , year= on PithPith's reviewarXiv
show 35 more
- [21] Bowen Jin and Hansi Zeng and Zhenrui Yue and Jinsung Yoon and Sercan O Arik and Dong Wang and Hamed Zamani and Jiawei Han , booktitle=. Search-R1: Training. 2025 , url= on Pith
- [22] Salton, G. and Wong, A. and Yang, C. S. , title =. Commun. ACM , month = nov, pages =. 1975 , issue_date =. doi:10.1145/361219.361220 , abstract = on PitharXiv
- [23] Salton, G. and Yang, C. S. and Yu, C. T. , title =. Journal of the American Society for Information Science , volume =. doi:https://doi.org/10.1002/asi.4630260106 , url =. https://asistdl.onlinelibrary.wiley.com/doi/pdf/10.1002/asi.4630260106 , abstract = on Pithdoi:10.1002/asi.4630260106
- [24] Morris, John X. and Rush, Alexander , booktitle =. Contextual Document Embeddings , url = on Pith
- [25] ReAct: Synergizing Reasoning and Acting in Language Models , author=. The Eleventh International Conference on Learning Representations , year= on Pith
- [26] Introducing Contextual Retrieval , author=. 2024 , howpublished= on Pith
- [27] Relevance feedback in information retrieval , author=. The SMART retrieval system: experiments in automatic document processing , year= on Pith
- [28] Lavrenko, Victor and Croft, W. Bruce , title =. Proceedings of the 24th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2001 , isbn =. doi:10.1145/383952.383972 , abstract = on PitharXiv
- [29] Su, Hongjin and Yen, Howard and Xia, Mengzhou and Shi, Weijia and Muennighoff, Niklas and Wang, Han-yu and Haisu, Liu and Shi, Quan and Siegel, Zachary and Tang, Michael and Sun, Ruoxi and Yoon, Jinsung and Arik, Sercan and Chen, Danqi and Yu, Tao , booktitle =. BRIGHT: A Real... on Pith
- [30] Lei, Fangyu and Chen, Jixuan and Ye, Yuxiao and Cao, Ruisheng and Shin, Dongchan and SU, Hongjin and Suo, Zhaoqing and Gao, Hongcheng and Hu, Wenjing and Yin, Pengcheng and Zhong, Victor and Xiong, Caiming and Sun, Ruoxi and Liu, Qian and Wang, Sida and Yu, Tao , booktitle =. ... on Pith
- [31] Sen, Jaydeep and Lei, Chuan and Quamar, Abdul and. ATHENA++: natural language querying for complex nested SQL queries , year =. Proc. VLDB Endow. , month = jul, pages =. doi:10.14778/3407790.3407858 , abstract = on PitharXiv
- [32] BEAVER: An Enterprise Benchmark for Text-to-SQL , author=. 2026 , eprint= on Pith
- [33] Romero Lauro, Quentin and Shankar, Shreya and Zeighami, Sepanta and Parameswaran, Aditya , title =. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems , articleno =. 2026 , isbn =. doi:10.1145/3772318.3790874 , abstract = on PitharXiv
- [34] Barnett, Scott and Kurniawan, Stefanus and Thudumu, Srikanth and Brannelly, Zach and Abdelrazek, Mohamed , title =. Proceedings of the IEEE/ACM 3rd International Conference on AI Engineering - Software Engineering for AI , pages =. 2024 , isbn =. doi:10.1145/3644815.3644945 , ... on PitharXiv
- [35] Gospodinov, Mitko and MacAvaney, Sean and Macdonald, Craig. Doc2Query--: When Less is More. Advances in Information Retrieval. 2023 on Pith
- [36] Na, Seung-Hoon and Kang, In-Su and Lee, Jong-Hyeok , title =. Proceedings of the IR Research, 30th European Conference on Advances in Information Retrieval , pages =. 2008 , isbn = on Pith
- [37] Lee, Katherine and Ippolito, Daphne and Nystrom, Andrew and Zhang, Chiyuan and Eck, Douglas and Callison-Burch, Chris and Carlini, Nicholas. Deduplicating Training Data Makes Language Models Better. Proceedings of the 60th Annual Meeting of the Association for Computational Li... on Pithdoi:10.18653/v1/2022.acl-long.577
- [38] Li, Jeffrey and Fang, Alex and Smyrnis, Georgios and Ivgi, Maor and Jordan, Matt and Gadre, Samir and Bansal, Hritik and Guha, Etash and Keh, Sedrick and Arora, Kushal and Garg, Saurabh and Xin, Rui and Muennighoff, Niklas and Heckel, Reinhard and Mercat, Jean and Chen, Mayee ... on Pith
- [39] BrowseComp-Plus: A More Fair and Transparent Evaluation Benchmark of Deep-Research Agent , author=. First Workshop on Multi-Turn Interactions in Large Language Models , year= on Pith
- [40] Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus Interaction , author=. 2026 , eprint= on Pith
- [41] GPT-4 Technical Report , author=. 2024 , eprint= on Pith
- [42] Wu, Di and Wang, Hongwei and Yu, Wenhao and Zhang, Yuwei and Chang, Kai-Wei and Yu, Dong , booktitle =. LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory , url = on Pith
- [43] LongMemEval-V2: Evaluating Long-Term Agent Memory Toward Experienced Colleagues , author=. 2026 , eprint= on Pith
- [44] Huang, Haoyu and Huang, Yongfeng and Junjie, Yang and Pan, Zhenyu and Chen, Yongqiang and Ma, Kaili and Chen, Hongzhi and Cheng, James. Retrieval-Augmented Generation with Hierarchical Knowledge. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025. doi:... on Pithdoi:10.18653/v1/2025.findings-emnlp.321
- [45] M-RAG: Semantic Key-Value Indexing for Retrieval-Augmented Generation , author=. 2026 , eprint= on Pith
- [46] Ouyang, Siru and Yan, Jun and Hsu, I-Hung and Chen, Yanfei and Jiang, Ke and Wang, Zifeng and Han, Rujun and Le, Long and Daruki, Samira and Tang, Xiangru and Tirumalashetty, Vishy and Lee, George and Rofouei, Mahsan and Lin, Hangfei and Han, Jiawei and Lee, Chen-Yu and Pfiste... on Pith
- [47] Persona2Web: Benchmarking Personalized Web Agents for Contextual Reasoning with User History , author=. Forty-third International Conference on Machine Learning , year= on Pith
- [48] gpt-oss-120b & gpt-oss-20b Model Card , author=. 2025 , eprint= on Pith
- [49] Kimi K2.5: Visual Agentic Intelligence , author=. 2026 , eprint= on Pith
- [50] Introducing GPT-5.4 Mini and Nano , year = on Pith
- [51] Gemini 3.7 Flash Model Card , institution =. 2026 , url = on Pith
- [52] Weller, Orion and Lo, Kyle and Wadden, David and Lawrie, Dawn and Van Durme, Benjamin and Cohan, Arman and Soldaini, Luca. When do Generative Query and Document Expansions Fail? A Comprehensive Study Across Methods, Retrievers, and Datasets. Findings of the Association for Com... on Pithdoi:10.18653/v1/2024.findings-eacl.134
- [53] Rulin Shao and Rui Qiao and Varsha Kishore and Niklas Muennighoff and Xi Victoria Lin and Daniela Rus and Bryan Kian Hsiang Low and Sewon Min and Wen-tau Yih and Pang Wei Koh and Luke Zettlemoyer , booktitle=. Reason. 2025 , url= on Pith
- [54] Group-Aware Adaptive Retrieval for Evidence Navigation , author=. 2026 , eprint= on Pith
- [55] SAGE: Benchmarking and Improving Retrieval for Deep Research Agents , author=. 2026 , eprint= on Pith
Cited by 1 Pith paper
Reviewed papers on Pith that cite this one. OpenAlex reports about 2 citations worldwide. The full citation record
- CoEvo-RSI: When Harnesses Evolve, Which Models Should Follow? cs.LG · Oct 2026
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.
No comments yet.
Sign in to comment