Pith. sign in

REVIEW 2 major objections 3 minor 3 cited by

MetaAgent: Toward Self-Evolving Agent via Tool Meta-Learning

T0 review · 2 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read According to its abstract, MetaAgent turns an agent's tool-use history into its own training data, improving performance on knowledge-discovery benchmarks without changing model parameters.

desk verdict The submission is an abstract for a MetaAgent AI paper attached to an unrelated math note; the empirical claims have no supporting evidence in the submitted text. read the letter →

arxiv 2508.00271 v2 pith:UFM7MDIA submitted 2025-08-01 cs.AI cs.CLcs.IR

classification cs.AIcs.CLcs.IR
keywords metatoollearningself-evolvingagentrouterknowledgediscoveryGAIAWebWalkerQABrowseCampbydoing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The abstract tries to establish that an agent can get better at knowledge-discovery tasks through use alone: starting from a minimal workflow, it asks for help, reflects on its answers, and folds the distilled lessons back into its context, without any parameter updates or post-training. The intended payoff is that this self-evolving agent outperforms workflow-based baselines and matches or exceeds end-to-end trained agents on GAIA, WebWalkerQA, and BrowseCamp, while meanwhile building its own tools and a persistent knowledge base from its tool-use history. As submitted, the paper's full text is an unrelated mathematics paper about Fourier integral operators; it contains none of the MetaAgent system, its experiments, or its results. The benchmark claims therefore cannot be checked against the submission.

What carries the argument

The central mechanism is meta tool learning, a continual loop in which the agent generates natural-language help requests, a dedicated tool router selects the best tool for each request, and self-reflection with answer verification distills actionable experience into short texts that are dynamically added to future task contexts. The same loop also organizes the agent's tool-use history into autonomously built tools and a persistent knowledge base. The whole mechanism is designed to refine reasoning and tool-use strategies without changing model parameters or running post-training.

What would settle it

Open the submitted PDF: the body is a Fourier integral operator paper that never mentions MetaAgent, the tool router, or the three benchmarks. If the promised code and experimental logs cannot be produced, or if the reported scores cannot be reproduced with the described method, the central claim fails.

Watch

Extended reading notes

Core claim

The paper's stated discovery is that learning-by-doing can be captured as a data process rather than a training process: MetaAgent starts with basic reasoning and help-seeking, routes each help request to the most suitable external tool, verifies its own answers, and distills each task's lessons into concise texts that become part of the context for future tasks. Over time its tool-use history is organized into in-house tools and a persistent knowledge base, so later tasks retrieve and integrate earlier experience. This 'meta tool learning' loop is claimed to be enough to outperform workflow-based baselines and to match or exceed end-to-end trained agents on GAIA, WebWalkerQA, and BrowseCamp. The submitted text does not contain this system or these results.

Load-bearing premise

The load-bearing premise is that the GAIA, WebWalkerQA, and BrowseCamp evaluation was actually run with the MetaAgent method described and compared fairly; the submitted text gives no way to check this, since its body is an unrelated mathematics paper.

Editorial extensions

If this is right

  • If the benchmark results are as reported, MetaAgent would match or beat end-to-end trained agents on general knowledge discovery without any post-training.
  • Workflow-based baselines would be a weaker setting than self-evolving context accumulation in these tasks.
  • A deployed agent could accumulate expertise during normal use, eliminating the need to retrain the underlying model after each new task type is encountered.

Reading between the lines

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

  • Because the submitted text is an unrelated math paper, the benchmark numbers in the abstract should be treated as unverified until a matching manuscript, code, and experimental logs are made available.
  • If a complete MetaAgent paper exists elsewhere, the decisive scalability question it would need to answer is whether the experience texts appended to the context eventually crowd out task-relevant information; the abstract does not address this.
  • The claimed comparison to end-to-end trained agents would only be meaningful if the base model, tool set, and test-time compute budget were identical across conditions; nothing in the submission establishes that.
  • A plausible testable extension would be to ablate the knowledge base and in-house tools while keeping the reflection loop, to see how much of the reported gain comes from tool building rather than from context distillation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 3 minor

Summary. The manuscript as submitted pairs an abstract claiming a new AI agent system, MetaAgent, with a full text that is an unrelated mathematics paper. The abstract describes a tool meta-learning paradigm, self-reflection, answer verification, and evaluations on GAIA, WebWalkerQA, and BrowseCamp, asserting that MetaAgent consistently outperforms workflow-based baselines and matches or exceeds end-to-end trained agents. The body, however, is a Fourier integral operators note internally labeled arXiv:2508.00279v2 [math.CA], presenting an L^4 bound for a family of operators. It contains no description of MetaAgent, no tool router, no meta tool learning, no benchmark results, and no algorithmic or experimental content that could support the abstract. The paper's actual content is therefore a mathematics paper with no connection to the claimed AI contribution.

Significance. If the abstract were supported by the submitted text, the claimed result would be significant: it would demonstrate that an agent can improve through its own tool-use history without parameter updates or post-training, and would provide empirical evidence on three knowledge discovery benchmarks. However, the submitted artifact provides no such evidence. The body is a self-contained mathematics note with its own theorem and proof, and the abstract's empirical claims are entirely unverified. No machine-checked proofs, reproducible code, experimental tables, or falsifiable predictions appear in connection with MetaAgent. The mismatch is total, so the paper cannot currently be assessed for scientific soundness. If the correct MetaAgent manuscript exists, it was not the artifact under review and would need a fresh review.

major comments (2)
  1. [Full text vs. abstract] The central claim of the abstract, namely that MetaAgent consistently outperforms workflow-based baselines and matches or exceeds end-to-end trained agents on GAIA, WebWalkerQA, and BrowseCamp, is not supported anywhere in the submitted full text. The body begins with '1. Introduction' to a mathematics paper defining Fourier integral operators and proving an L^4 estimate, and it is internally labeled 'arXiv:2508.00279v2 [math.CA] 9 Mar 2026'. There is no description of MetaAgent, no tool router, no meta tool learning procedure, no self-reflection or verification mechanism, and no benchmark evaluation. The abstract's claims are thus unverifiable from the submitted artifact.
  2. [Full text, Sections 1-2] Because the body contains the mathematics paper rather than the MetaAgent method, there are no experimental tables, ablations, statistical analyses, or baseline comparisons for the claimed benchmarks. The reader cannot assess the fairness of the comparisons to workflow-based baselines or end-to-end trained agents, nor the validity of the claimed performance. This is not a local omission but the complete absence of the evidence required for the paper's central contribution.
minor comments (3)
  1. [Abstract] The sentence 'We term this continual, data-driven process as \textit{meta tool learning}' is grammatically awkward; 'as' is unnecessary, and a comma after 'process' would improve readability.
  2. [Abstract] The phrase 'We provide our source codes' should be 'We provide our source code', as 'code' is a mass noun in this context.
  3. [Full text, displayed equations] The mathematics text contains extensive rendering artifacts, such as replacement characters and garbled expressions, which would make the proof difficult to follow even if it were the intended content of this submission.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the abstract's empirical claims have no derivation chain in the submitted body, which is an unrelated math note; unsupportedness is not circularity.

full rationale

The submitted artifact for arXiv:2508.00271 contains only the MetaAgent abstract followed by a mathematics note internally labeled 'arXiv:2508.00279v2 [math.CA] 9 Mar 2026', which proves an L^4 bound for Fourier integral operators (e.g., 'Let I be a compact interval in R and ψ a real valued function in C∞(I)... In this note we shall prove the following... ||S_λ^* f||_4 ≤ C_λ ||f||_4'). There is no description of MetaAgent, the tool router, meta tool learning, GAIA, WebWalkerQA, BrowseCamp, or any experimental table. The abstract's central claim—'MetaAgent consistently outperforms workflow-based baselines and matches or exceeds end-to-end trained agents'—is an empirical performance assertion, not a derived result. No equation, fitted parameter, self-citation, or uniqueness theorem appears that would make the claimed outcome equal to its input by construction. The absence of the reported experiments and method description is a serious verifiability and correctness defect, but under the hard rules circularity requires exhibiting a specific reduction in the paper's own reasoning; no such reduction exists in the visible text. Therefore the circularity score is 0.

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

No free parameters, axioms, or invented entities can be listed for the abstract because the manuscript body is unrelated and does not describe the method. The math note in the body does assume conditions (A.1) and (A.2), and the theorem assumes psi'' != 0, but those belong to a different paper and are not part of the MetaAgent claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MetaAgent: Toward Self-Evolving Agent via Tool Meta-Learning." pith.science (2026). https://pith.science/paper/UFM7MDIA

@misc{pith2026250800271,
  author       = {Pith},
  title        = {Pith review of: MetaAgent: Toward Self-Evolving Agent via Tool Meta-Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UFM7MDIA}},
  note         = {Machine review of arXiv:2508.00271}
}
read the original abstract

In this work, we propose MetaAgent, an agentic paradigm inspired by the principle of learning-by-doing, where expertise is developed through hands-on practice and continual self-improvement. MetaAgent starts with a minimal workflow, equipped only with basic reasoning and adaptive help-seeking abilities. When a knowledge gap is encountered, MetaAgent generates natural language help requests, which are routed to the most suitable external tool by a dedicated tool router. As MetaAgent solves tasks, it continually conducts self-reflection and answer verification, distilling actionable experience into concise texts that are dynamically incorporated into future task contexts. Besides, MetaAgent autonomously builds in-house tools and a persistent knowledge base by organizing its tool-use history, further enhancing its ability to retrieve and integrate relevant information We term this continual, data-driven process as \textit{meta tool learning}, through which MetaAgent incrementally refines its reasoning and tool-use strategies, without changing model parameters or requiring further post-training. Evaluated on challenging knowledge discovery benchmarks, including GAIA, WebWalkerQA, and BrowseCamp, MetaAgent consistently outperforms workflow-based baselines and matches or exceeds end-to-end trained agents, demonstrating the promise of self-evolving agentic systems for robust, general-purpose knowledge discovery. We provide our source codes in https://github.com/qhjqhj00/MetaAgent.

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

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

  1. Escaping the Self-Confirmation Trap: An Execute-Distill-Verify Paradigm for Agentic Experience Learning

    cs.CL 2026-06 unverdicted novelty 6.0 of 10

    EDV decouples execution, distillation by a third-party agent, and consensus verification to filter erroneous trajectories in LLM agent experience learning, outperforming baselines on tau2-bench, Mind2Web, and MMTB.

  2. MetaPS: Adaptive Programmatic Strategy Selection for Market Agents

    cs.AI 2026-06 unverdicted novelty 6.0 of 10

    MetaPS trains models via simulation rollouts to select from programmatic strategy libraries for market agents, yielding better performance than fixed or direct LLM baselines across model sizes.

  3. Self-Improvements in Modern Agentic Systems: A Survey

    cs.AI 2026-07 conditional novelty 5.0 of 10

    Self-improving agents are classified by what they update — foundation-model weights or the surrounding scaffold — and by the signal that drives the update, under a single formal operator.

Reference graph

Works this paper leans on

43 extracted references · 6 canonical work pages · cited by 3 Pith papers

  1. [1]

    Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation, 2023

    Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu. Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation, 2023

  2. [2]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning

    DeepSeek - AI. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. CoRR, abs/2501.12948, 2025. doi:10.48550/ARXIV.2501.12948. URL https://doi.org/10.48550/arXiv.2501.12948

  3. [3]

    Tool-star: Empowering llm-brained multi-tool reasoner via reinforcement learning

    Guanting Dong, Yifei Chen, Xiaoxi Li, Jiajie Jin, Hongjin Qian, Yutao Zhu, Hangyu Mao, Guorui Zhou, Zhicheng Dou, and Ji-Rong Wen. Tool-star: Empowering llm-brained multi-tool reasoner via reinforcement learning. arXiv preprint arXiv:2505.16410, 2025

  4. [4]

    Retrieval-augmented generation for large language models: A survey, 2024

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Qianyu Guo, Meng Wang, and Haofen Wang. Retrieval-augmented generation for large language models: A survey, 2024

  5. [5]

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities, 2025

    Google Gemini Team. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities, 2025. URL https://arxiv.org/abs/2507.06261

  6. [6]

    Gemini: A family of highly capable multimodal models

    Google. Gemini: A family of highly capable multimodal models. https://goo.gle/GeminiPaper, 2023

  7. [7]

    Qwen2.5 technical report, 2025

    Qwen Group. Qwen2.5 technical report, 2025. URL https://arxiv.org/abs/2412.15115

  8. [8]

    An empirical study on reinforcement learning for reasoning-search interleaved llm agents

    Bowen Jin, Jinsung Yoon, Priyanka Kargupta, Sercan O Arik, and Jiawei Han. An empirical study on reinforcement learning for reasoning-search interleaved llm agents. arXiv preprint arXiv:2505.15117, 2025 a

Show all 43 references
  1. [9]

    Search-r1: Training llms to reason and leverage search engines with reinforcement learning

    Bowen Jin, Hansi Zeng, Zhenrui Yue, Dong Wang, Hamed Zamani, and Jiawei Han. Search-r1: Training llms to reason and leverage search engines with reinforcement learning. CoRR, abs/2503.09516, 2025 b . doi:10.48550/ARXIV.2503.09516. URL https://doi.org/10.48550/arXiv.2503.09516

  2. [10]

    u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt\

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K\" u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt\" a schel, Sebastian Riedel, and Douwe Kiela. R etrieval- A ugmented G eneration for knowledge-intensive NLP tasks. In Advanc...

  3. [11]

    Search-o1: Agentic search-enhanced large reasoning models

    Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, Peitian Zhang, and Zhicheng Dou. Search-o1: Agentic search-enhanced large reasoning models. CoRR, abs/2501.05366, 2025 a . doi:10.48550/ARXIV.2501.05366. URL https://doi.org/10.48550/arXiv.2501.05366

  4. [12]

    Webthinker: Empowering large reasoning models with deep research capability

    Xiaoxi Li, Jiajie Jin, Guanting Dong, Hongjin Qian, Yutao Zhu, Yongkang Wu, Ji-Rong Wen, and Zhicheng Dou. Webthinker: Empowering large reasoning models with deep research capability. arXiv preprint arXiv:2504.21776, 2025 b

  5. [13]

    Towards ai search paradigm

    Yuchen Li, Hengyi Cai, Rui Kong, Xinran Chen, Jiamin Chen, Jun Yang, Haojie Zhang, Jiayi Li, Jiayi Wu, Yiqun Chen, et al. Towards ai search paradigm. arXiv preprint arXiv:2506.17188, 2025 c

  6. [14]

    Gaia: a benchmark for general ai assistants

    Gr \'e goire Mialon, Cl \'e mentine Fourrier, Thomas Wolf, Yann LeCun, and Thomas Scialom. Gaia: a benchmark for general ai assistants. In The Twelfth International Conference on Learning Representations, 2023

  7. [15]

    Webgpt: Browser-assisted question-answering with human feedback

    Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, et al. Webgpt: Browser-assisted question-answering with human feedback. arXiv preprint arXiv:2112.09332, 2021

  8. [16]

    Gpt-4 technical report

    OpenAI. Gpt-4 technical report. https://cdn.openai.com/papers/gpt-4.pdf, 2023

  9. [17]

    Introducing deep research, 2025

    OpenAI. Introducing deep research, 2025. URL https://openai.com/index/introducing-deep-research/

  10. [18]

    Training language models to follow instructions with human feedback

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35: 0 2...

  11. [19]

    Feng, Haoran Zhao, Michael Yu, Varun Gangal, Chelsea Zou, Zihan Wang, Jessica P

    Long Phan, Alice Gatti, Ziwen Han, Nathaniel Li, Josephina Hu, Hugh Zhang, Sean Shi, Michael Choi, Anish Agrawal, Arnav Chopra, Adam Khoja, Ryan Kim, Jason Hausenloy, Oliver Zhang, Mantas Mazeika, Daron Anderson, Tung Nguyen, Mobeen Mahmood, Fiona Feng, Steven Y. Feng, Haoran ...

  12. [20]

    Hawkbench: Investigating resilience of rag methods on stratified information-seeking tasks

    Hongjin Qian, Zheng Liu, Chao Gao, Yankai Wang, Defu Lian, and Zhicheng Dou. Hawkbench: Investigating resilience of rag methods on stratified information-seeking tasks. arXiv preprint arXiv:2502.13465, 2025

  13. [21]

    Metacognitive theories

    Gregory Schraw and David Moshman. Metacognitive theories. Educational psychology review, 7 0 (4): 0 351--371, 1995

  14. [22]

    Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy

    Zhihong Shao, Yeyun Gong, Yelong Shen, Minlie Huang, Nan Duan, and Weizhu Chen. Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy. arXiv preprint arXiv:2305.15294, 2023

  15. [23]

    Search and refine during think: Autonomous retrieval-augmented reasoning of llms, 2025

    Yaorui Shi, Sihang Li, Chang Wu, Zhiyuan Liu, Junfeng Fang, Hengxing Cai, An Zhang, and Xiang Wang. Search and refine during think: Autonomous retrieval-augmented reasoning of llms, 2025. URL https://arxiv.org/abs/2505.11277

  16. [24]

    Scaling llm test-time compute optimally can be more effective than scaling model parameters

    Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. Scaling llm test-time compute optimally can be more effective than scaling model parameters. arXiv preprint arXiv:2408.03314, 2024

  17. [25]

    Coding agents with multimodal browsing are generalist problem solvers

    Aditya Bharat Soni, Boxuan Li, Xingyao Wang, Valerie Chen, and Graham Neubig. Coding agents with multimodal browsing are generalist problem solvers. arXiv preprint arXiv:2506.03011, 2025

  18. [26]

    Acting less is reasoning more! teaching model to act efficiently, 2025 a

    Hongru Wang, Cheng Qian, Wanjun Zhong, Xiusi Chen, Jiahao Qiu, Shijue Huang, Bowen Jin, Mengdi Wang, Kam-Fai Wong, and Heng Ji. Acting less is reasoning more! teaching model to act efficiently, 2025 a . URL https://arxiv.org/abs/2504.14870

  19. [27]

    A survey on large language model based autonomous agents

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, Wayne Xin Zhao, Zhewei Wei, and Jirong Wen. A survey on large language model based autonomous agents. Frontiers Comput. Sci., 18 0 (6): 0 186345, 2024. doi:10....

  20. [28]

    Chain-of-retrieval augmented generation, 2025 b

    Liang Wang, Haonan Chen, Nan Yang, Xiaolong Huang, Zhicheng Dou, and Furu Wei. Chain-of-retrieval augmented generation, 2025 b . URL https://arxiv.org/abs/2501.14342

  21. [29]

    Browsecomp: A simple yet challenging benchmark for browsing agents

    Jason Wei, Zhiqing Sun, Spencer Papay, Scott McKinney, Jeffrey Han, Isa Fulford, Hyung Won Chung, Alex Tachard Passos, William Fedus, and Amelia Glaese. Browsecomp: A simple yet challenging benchmark for browsing agents. arXiv preprint arXiv:2504.12516, 2025

  22. [30]

    Webdancer: Towards autonomous information seeking agency, 2025 a

    Jialong Wu, Baixuan Li, Runnan Fang, Wenbiao Yin, Liwen Zhang, Zhengwei Tao, Dingchu Zhang, Zekun Xi, Gang Fu, Yong Jiang, Pengjun Xie, Fei Huang, and Jingren Zhou. Webdancer: Towards autonomous information seeking agency, 2025 a . URL https://arxiv.org/abs/2505.22648

  23. [31]

    Webwalker: Benchmarking llms in web traversal

    Jialong Wu, Wenbiao Yin, Yong Jiang, Zhenglin Wang, Zekun Xi, Runnan Fang, Linhai Zhang, Yulan He, Deyu Zhou, Pengjun Xie, et al. Webwalker: Benchmarking llms in web traversal. arXiv preprint arXiv:2501.07572, 2025 b

  24. [32]

    Agentic reasoning: Reasoning llms with tools for the deep research

    Junde Wu, Jiayuan Zhu, and Yuyuan Liu. Agentic reasoning: Reasoning llms with tools for the deep research. arXiv preprint arXiv:2502.04644, 2025 c

  25. [33]

    React: Synergizing reasoning and acting in language models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR), 2023

  26. [34]

    Agentic information retrieval

    Weinan Zhang, Junwei Liao, Ning Li, Kounianhua Du, and Jianghao Lin. Agentic information retrieval. arXiv preprint arXiv:2410.09713, 2024

  27. [35]

    Process vs

    Wenlin Zhang, Xiangyang Li, Kuicai Dong, Yichao Wang, Pengyue Jia, Xiaopeng Li, Yingyi Zhang, Derong Xu, Zhaocheng Du, Huifeng Guo, Ruiming Tang, and Xiangyu Zhao. Process vs. outcome reward: Which is better for agentic rag reinforcement learning, 2025. URL https://arxiv.org/a...

  28. [36]

    Qiu, and Lili Qiu

    Siyun Zhao, Yuqing Yang, Zilong Wang, Zhiyuan He, Luna K. Qiu, and Lili Qiu. Retrieval augmented generation (rag) and beyond: A comprehensive survey on how to make your llms use external data more wisely, 2024 a . URL https://arxiv.org/abs/2409.14924

  29. [37]

    A survey of large language models, 2024 b

    Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, Yifan Du, Chen Yang, Yushuo Chen, Zhipeng Chen, Jinhao Jiang, Ruiyang Ren, Yifan Li, Xinyu Tang, Zikang Liu, Peiyu Liu, Jian-Yun Nie, and Ji-Rong W...

  30. [38]

    Deepresearcher: Scaling deep research via reinforcement learning in real-world environments, 2025

    Yuxiang Zheng, Dayuan Fu, Xiangkun Hu, Xiaojie Cai, Lyumanshan Ye, Pengrui Lu, and Pengfei Liu. Deepresearcher: Scaling deep research via reinforcement learning in real-world environments, 2025. URL https://arxiv.org/abs/2504.03160

  31. [39]

    Large language models for information retrieval: A survey, 2024

    Yutao Zhu, Huaying Yuan, Shuting Wang, Jiongnan Liu, Wenhan Liu, Chenlong Deng, Haonan Chen, Zhicheng Dou, and Ji-Rong Wen. Large language models for information retrieval: A survey, 2024. URL https://arxiv.org/abs/2308.07107

  32. [40]

    write newline

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

  33. [41]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  34. [42]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  35. [43]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

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