pith. sign in

arxiv: 2605.29612 · v1 · pith:IZX57RXLnew · submitted 2026-05-28 · 💻 cs.MA · cs.CL

CONCAT: Consensus- and Confidence-Driven Ad Hoc Teaming for Efficient LLM-Based Multi-Agent Systems

Ziyang Ma , Dingyi Zhang , Sichu Liang , Jiajia Chu , Pengfei Xia , Hui Zang , Deyu Zhou This is my paper

Pith reviewed 2026-06-29 00:10 UTC · model grok-4.3

classification 💻 cs.MA cs.CL
keywords multi-agent systemslarge language modelsad hoc teamingconsensus clusteringconfidence estimationtraining-free methodscommunication pruningefficiency optimization
0
0 comments X

The pith

By clustering agents on their answers and pruning talks via a confidence-based heuristic, multi-agent LLM systems cut latency in half without any training.

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

The paper tries to establish that full communication among LLM agents is unnecessary and wasteful for many tasks. Instead, agents can be grouped by their initial answers, leaders chosen by confidence, and low-benefit pairs identified by a simple prediction rule so that only useful conversations remain. If correct, this removes the need for expensive per-task training while still solving complex problems at lower computational cost. A sympathetic reader would care because current multi-agent setups either waste resources on constant chatter or lock themselves to narrow domains through retraining. The result points toward more scalable, general-purpose agent teams that adapt on the fly.

Core claim

CONCAT clusters agents according to their initial answers, selects cluster leaders by reported confidence, and applies a Theory of Mind heuristic to forecast collaboration benefits between every pair of leaders from their answers and scores. Communications predicted to deliver little benefit are then evicted, producing an ad hoc network that delivers up to 2.02 times higher accuracy-per-latency than full LLM-Debate while cutting average latency by 50.1 percent on Qwen2.5-14B-Instruct, all without task-specific training and outperforming certain training-aware baselines on three benchmarks.

What carries the argument

The Theory of Mind heuristic that estimates pairwise collaboration benefits from leaders' answers and confidence values to decide which communications to keep.

If this is right

  • Accuracy divided by latency reaches up to 2.02 times the value of LLM-Debate across three models and three benchmarks.
  • Average latency drops by 50.1 percent on Qwen2.5-14B-Instruct while performance stays competitive.
  • The method beats training-aware approaches such as AgentDropout on the reported efficiency metric.
  • No task-specific training is required, preserving applicability across different LLMs and benchmarks.

Where Pith is reading between the lines

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

  • The same clustering-plus-pruning pattern could be tested on agent teams that include non-LLM components such as symbolic planners.
  • If the heuristic remains stable when agent count grows, the approach might support real-time team reorganization in long-running multi-agent workflows.
  • The reported gains suggest that similar confidence signals might reduce message volume in other distributed reasoning systems that currently rely on complete graphs.

Load-bearing premise

The heuristic function based on the Theory of Mind accurately predicts the collaboration benefits between every two leaders according to their answers and confidence.

What would settle it

An experiment in which teams formed by the heuristic show lower accuracy than either the full-communication baseline or teams formed by random eviction of the same number of links.

Figures

Figures reproduced from arXiv: 2605.29612 by Deyu Zhou, Dingyi Zhang, Hui Zang, Jiajia Chu, Pengfei Xia, Sichu Liang, Ziyang Ma.

Figure 1
Figure 1. Figure 1: Statistics of two types of answer changes [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Collaboration outcome distribution on GSM8K using LLM-Debate (Du et al., 2024) with two, three, four, and five agents. Each agent pair is categorized by its answer transition across rounds: Wrong→Correct, Correct→Correct, Wrong→Wrong, or Correct→Wrong. Observation 1: Referencing other agents has negative or neutral impacts more frequently than correcting errors. 2.3 Observation 2: Predictability of Collabo… view at source ↗
Figure 4
Figure 4. Figure 4: Overview of CONCAT. CONCAT operates through three phases: (1) Initialization, where each agent independently generates an answer; (2) Ad Hoc Teaming, where agents are grouped by consensus clus￾tering and leader selection, followed by benefit-driven edge pruning to construct a sparse communication topol￾ogy, repeated for (m − 1) rounds; (3) Final Answer Aggregation, where an LLM synthesizer aggregates the a… view at source ↗
Figure 3
Figure 3. Figure 3: ROC-AUC of dissent strength (dj→k = c¯j · (1 − agreejk)) for predicting helpful collaboration (Wrong→Correct) across 2–5 agent configurations on GSM8K and MMLU. All results computed on LLM￾Debate (Du et al., 2024) based on Llama-3-8B-Instruct. Observation 2: Collaboration effectiveness is predictable from answer similarity and agent confidence scores, enabling training-free and principled edge pruning. 3 M… view at source ↗
Figure 5
Figure 5. Figure 5: Efficiency comparison of multi-agent methods [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: GSM8K Answer Aggregation Prompt System Prompt: You are the top decision-maker. Good at analyzing and summarizing mathematical problems, judging and summarizing other people's solutions, and giving final answers to math problems. You will be given a math problem, analysis and code from other agents. Please find the most reliable answer based on the analysis and results of other agents. Give reasons for maki… view at source ↗
Figure 8
Figure 8. Figure 8: HumanEval Answer Aggregation Prompt System Prompt: You are the top decision-maker and are good at analyzing and summarizing other people's opinions, finding errors and giving final answers. And you are an AI that only responds with only python code. You will be given a function signature and its docstring by the user. You may be given the overall code design, algorithm framework, code implementation or tes… view at source ↗
read the original abstract

Although large language model (LLM) based multi-agent systems (MAS) show their capability to solve complex tasks and achieve higher performance over single agent systems, they lead to huge computational overheads because of heavy communication between agents. Previous research has made efforts to train a sparse multi-agent graph or fine-tune a planner to orchestrate the workflow better. However, such extra training processes introduce computational costs and limit MAS to specific domains, therefore compromising their generalizability. In this paper, we propose CONCAT, a training-free multi-agent collaboration framework based on CONsensus and Confidence-driven Ad hoc Teaming to efficiently organize agent interactions. Specifically, agents are clustered based on their initial answers, and leaders of each cluster are selected based on the agents' confidence. Then, a heuristic function based on the Theory of Mind is designed to predict the collaboration benefits between every two leaders according to their answers and confidence. Finally, an ad hoc multi-agent network is organized after evicting a percentage of communications based on the predicted benefits. Experiments across three LLMs and three benchmarks show that CONCAT achieves up to 2.02x higher efficiency (accuracy/latency ratio) than LLM-Debate and outperforms training-aware methods such as AgentDropout, while reducing average latency by 50.1% on Qwen2.5-14B-Instruct, without any task-specific training.

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.

Referee Report

2 major / 1 minor

Summary. The paper proposes CONCAT, a training-free multi-agent collaboration framework for LLM-based systems. Agents are clustered based on initial answers and leaders selected by confidence; a Theory of Mind heuristic then predicts pairwise collaboration benefits to evict a percentage of communications and form an ad hoc network. Experiments across three LLMs and three benchmarks report up to 2.02x higher accuracy/latency efficiency than LLM-Debate, outperformance versus training-aware baselines such as AgentDropout, and a 50.1% average latency reduction on Qwen2.5-14B-Instruct.

Significance. If the reported efficiency gains hold and the ToM heuristic is shown to be the operative mechanism, the work would be significant for enabling generalizable, low-overhead multi-agent LLM systems without task-specific training or fine-tuning costs. The training-free design and concrete latency/accuracy trade-off improvements address a practical bottleneck in current MAS deployments.

major comments (2)
  1. [Abstract / Experiments] Abstract and Experiments section: the central efficiency claims (2.02x ratio, 50.1% latency reduction) are presented without accompanying details on baseline implementations, number of runs, error bars, or statistical tests. These elements are load-bearing for verifying that the observed gains exceed those from generic communication reduction.
  2. [Method] Method description: the ToM-based heuristic for predicting collaboration benefits between leader pairs is the key innovation enabling selective eviction. Without an ablation that isolates this heuristic from simpler clustering or random eviction, it remains unclear whether the reported gains are attributable to the specific benefit-prediction mechanism rather than reduced communication volume alone.
minor comments (1)
  1. [Method] Notation for the heuristic function and clustering procedure could be formalized with explicit equations or pseudocode to improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and outline the revisions we will make to the manuscript.

read point-by-point responses

  1. Referee: [Abstract / Experiments] Abstract and Experiments section: the central efficiency claims (2.02x ratio, 50.1% latency reduction) are presented without accompanying details on baseline implementations, number of runs, error bars, or statistical tests. These elements are load-bearing for verifying that the observed gains exceed those from generic communication reduction.

    Authors: We agree that additional experimental details are needed to substantiate the efficiency claims. In the revised manuscript, we will expand the Experiments section to specify baseline implementations (reproduced following the original papers), report results averaged over 5 independent runs with standard error bars, and include statistical tests (paired t-tests) comparing CONCAT against baselines. These additions will help confirm that gains are not attributable solely to generic communication reduction. We will also note these details briefly in the abstract if space allows. revision: yes

  2. Referee: [Method] Method description: the ToM-based heuristic for predicting collaboration benefits between leader pairs is the key innovation enabling selective eviction. Without an ablation that isolates this heuristic from simpler clustering or random eviction, it remains unclear whether the reported gains are attributable to the specific benefit-prediction mechanism rather than reduced communication volume alone.

    Authors: We acknowledge that an ablation isolating the Theory of Mind heuristic would strengthen the paper. We will add this analysis in the revised version, including comparisons of CONCAT against (i) a variant using random eviction at the same communication reduction rate and (ii) a variant using only initial clustering without the ToM benefit prediction. This will clarify the contribution of the heuristic beyond volume reduction alone. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes CONCAT as a training-free heuristic that clusters agents by initial answers, selects leaders by confidence, applies a Theory of Mind-based benefit prediction to evict communications, and organizes an ad hoc network. No equations, fitted parameters, or self-citations appear in the provided text that would reduce any claimed prediction or result to the method's own inputs by construction. The efficiency gains are presented as outcomes of experimental evaluation against baselines rather than definitional or self-referential equivalences. The derivation chain remains independent of the patterns that trigger circularity flags.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities are specified in the provided text.

pith-pipeline@v0.9.1-grok · 5797 in / 1189 out tokens · 21583 ms · 2026-06-29T00:10:50.240208+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 24 canonical work pages · 13 internal anchors

  1. [1]

    Baker, Julian Jara-Ettinger, Rebecca Saxe, and Joshua B

    Chris L. Baker, Julian Jara-Ettinger, Rebecca Saxe, and Joshua B. Tenenbaum. 2017. https://doi.org/10.1038/s41562-017-0064 Rational quantitative attribution of beliefs, desires and percepts in human mentalizing . Nature Human Behaviour, 1:0064

    work page doi:10.1038/s41562-017-0064 2017

  2. [2]

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, and 39 others. 2021. https://doi.org/10.48550/arXiv.2107.03374 Evaluating L...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2107.03374 2021

  3. [3]

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, and 1 others. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  4. [4]

    Logan Cross, Violet Xiang, Agam Bhatia, Daniel Yamins, and Nick Haber. 2025. Hypothetical minds: Scaffolding theory of mind for multi-agent tasks with large language models. In International Conference on Learning Representations, volume 2025, pages 6507--6546

    2025

  5. [5]

    Yufan Dang, Chen Qian, Xueheng Luo, Jingru Fan, Zihao Xie, Ruijie Shi, Weize Chen, Cheng Yang, Xiaoyin Che, Ye Tian, Xuantang Xiong, Lei Han, Zhiyuan Liu, and Maosong Sun. 2025. https://doi.org/10.48550/arXiv.2505.19591 Multi- Agent Collaboration via Evolving Orchestration . Preprint, arXiv:2505.19591

    work page doi:10.48550/arxiv.2505.19591 2025

  6. [6]

    DeepSeek-AI , Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, and 181 others. 2025. https://doi.org/10.48550/arXiv.2501.12948 DeepSeek-R1 : Incentivizing Reasoning Capability in LLMs via Reinfo...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2501.12948 2025

  7. [7]

    Tenenbaum, and Igor Mordatch

    Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, and Igor Mordatch. 2024. Improving Factuality and Reasoning in Language Models through Multiagent Debate . In Forty-First International Conference on Machine Learning

    2024

  8. [8]

    Adam Fourney, Gagan Bansal, Hussein Mozannar, Cheng Tan, Eduardo Salinas, Erkang, Zhu, Friederike Niedtner, Grace Proebsting, Griffin Bassman, Jack Gerrits, Jacob Alber, Peter Chang, Ricky Loynd, Robert West, Victor Dibia, Ahmed Awadallah, Ece Kamar, Rafah Hosn, and Saleema Amershi. 2024. https://doi.org/10.48550/arXiv.2411.04468 Magentic- One : A General...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2411.04468 2024

  9. [9]

    Carlin, Hal S

    Andrew Gelman, John B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, and Donald B. Rubin. 2013. Bayesian Data Analysis, 3rd edition. CRC Press

    2013

  10. [10]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle , Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, and 542 others. 2024. https://doi.org/10.48550/arXiv.2407.21783 T...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2407.21783 2024

  11. [11]

    Large Language Model based Multi-Agents: A Survey of Progress and Challenges

    Taicheng Guo, Xiuying Chen, Yaqi Wang, Ruidi Chang, Shichao Pei, Nitesh V. Chawla, Olaf Wiest, and Xiangliang Zhang. 2024. https://arxiv.org/abs/2402.01680 Large language model based multi-agents: A survey of progress and challenges . arXiv preprint arXiv:2402.01680

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  12. [12]

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2020. Measuring Massive Multitask Language Understanding . In International Conference on Learning Representations

    2020

  13. [13]

    Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, and J \"u rgen Schmidhuber. 2023. MetaGPT : Meta Programming for A Multi-Agent Collaborative Framework . In The Twelfth International Conference on Learning Representations

    2023

  14. [14]

    Ronald A. Howard. 1966. https://doi.org/10.1109/TSSC.1966.300074 Information value theory . IEEE Transactions on Systems Science and Cybernetics, 2(1):22--26

    work page doi:10.1109/tssc.1966.300074 1966

  15. [15]

    Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, and 1 others. 2024. Openai o1 system card. arXiv preprint arXiv:2412.16720

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  16. [16]

    Adam Kostka and Jaros aw A Chudziak. 2025. Evaluating theory of mind and internal beliefs in llm-based multi-agent systems. In International Conference on Computational Collective Intelligence, pages 18--32. Springer

    2025

  17. [17]

    Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the 29th symposium on operating systems principles, pages 611--626

    2023

  18. [18]

    Huao Li, Yu Chong, Simon Stepputtis, Joseph Campbell, Dana Hughes, Michael Lewis, and Katia Sycara. 2023. https://arxiv.org/abs/2310.10701 Theory of mind for multi-agent collaboration via large language models . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP)

    work page arXiv 2023

  19. [19]

    Xiaoxi Li, Jiajie Jin, Guanting Dong, Hongjin Qian, Yongkang Wu, Ji-Rong Wen, Yutao Zhu, and Zhicheng Dou. 2025. https://doi.org/10.48550/arXiv.2504.21776 WebThinker : Empowering Large Reasoning Models with Deep Research Capability . Preprint, arXiv:2504.21776

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2504.21776 2025

  20. [20]

    Gr \'e goire Mialon, Cl \'e mentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. 2023. https://doi.org/10.48550/arXiv.2311.12983 GAIA : A benchmark for General AI Assistants . Preprint, arXiv:2311.12983

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2311.12983 2023

  21. [21]

    Chunjiang Mu, Ya Zeng, Qiaosheng Zhang, Kun Shao, Chen Chu, Hao Guo, Danyang Jia, Zhen Wang, and Shuyue Hu. 2026. Adaptive theory of mind for llm-based multi-agent coordination. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 29608--29616

    2026

  22. [22]

    David Premack and Guy Woodruff. 1978. https://doi.org/10.1017/S0140525X00076512 Does the chimpanzee have a theory of mind? Behavioral and Brain Sciences, 1(4):515--526

    work page doi:10.1017/s0140525x00076512 1978

  23. [23]

    Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, Juyuan Xu, Dahai Li, Zhiyuan Liu, and Maosong Sun. 2024. https://arxiv.org/abs/2307.07924 ChatDev : Communicative agents for software development . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  24. [24]

    Haojun Shi, Suyu Ye, Xinyu Fang, Chuanyang Jin, Leyla Isik, Yen-Ling Kuo, and Tianmin Shu. 2025. Muma-tom: Multi-modal multi-agent theory of mind. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 1510--1519

    2025

  25. [25]

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. https://arxiv.org/abs/2303.11366 Reflexion: Language agents with verbal reinforcement learning . In Advances in Neural Information Processing Systems (NeurIPS)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  26. [26]

    Chenglei Si, Zhe Gan, Zhengyuan Yang, Shuohang Wang, Jianfeng Wang, Jordan Lee Boyd-Graber , and Lijuan Wang. 2022. Prompting GPT-3 To Be Reliable . In The Eleventh International Conference on Learning Representations

    2022

  27. [27]

    Amos Tversky and Daniel Kahneman. 1974. https://doi.org/10.1126/science.185.4157.1124 Judgment under uncertainty: Heuristics and biases . Science, 185(4157):1124--1131

    work page doi:10.1126/science.185.4157.1124 1974

  28. [28]

    Vllm-Project. 2025. https://github.com/vllm-project/vllm-ascend Vllm-ascend

    2025

  29. [29]

    Le, Ed H

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. Self- Consistency Improves Chain of Thought Reasoning in Language Models . In The Eleventh International Conference on Learning Representations

    2022

  30. [30]

    Wong, and Rui Wang

    Yiming Wang, Pei Zhang, Baosong Yang, Derek F. Wong, and Rui Wang. 2024. Latent Space Chain-of-Embedding Enables Output-free LLM Self-Evaluation . In The Thirteenth International Conference on Learning Representations

    2024

  31. [31]

    Zhexuan Wang, Yutong Wang, Xuebo Liu, Liang Ding, Miao Zhang, Jie Liu, and Min Zhang. 2025 a . https://doi.org/10.18653/v1/2025.acl-long.1170 AgentDropout : Dynamic Agent Elimination for Token-Efficient and High-Performance LLM-Based Multi-Agent Collaboration . In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics ( Vo...

    work page doi:10.18653/v1/2025.acl-long.1170 2025

  32. [32]

    Zihao Wang, Shaofei Cai, Anji Liu, Yonggang Jin, Jinbing Hou, Bowei Zhang, Haowei Lin, Zhaofeng He, Zilong Zheng, Yaodong Yang, Xiaojian Ma, and Yitao Liang. 2025 b . https://doi.org/10.1109/TPAMI.2024.3511593 JARVIS-1 : Open-World Multi-Task Agents With Memory-Augmented Multimodal Language Models . IEEE Transactions on Pattern Analysis and Machine Intell...

    work page doi:10.1109/tpami.2024.3511593 2025

  33. [33]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824--24837

    2022

  34. [34]

    Zhiyuan Weng, Guikun Chen, and Wenguan Wang. 2024. Do as We Do , Not as You Think : The Conformity of Large Language Models . In The Thirteenth International Conference on Learning Representations

    2024

  35. [35]

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, and 40 others. 2024. Qwen2 technical report. arXiv preprint arXiv:2407.10671

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  36. [36]

    Hancheng Ye, Zhengqi Gao, Mingyuan Ma, Qinsi Wang, Yuzhe Fu, Ming-Yu Chung, Yueqian Lin, Zhijian Liu, Jianyi Zhang, Danyang Zhuo, and Yiran Chen. 2025 a . KVCOMM : Online Cross-context KV-cache Communication for Efficient LLM-based Multi-agent Systems . In The Thirty-ninth Annual Conference on Neural Information Processing Systems

    2025

  37. [37]

    Rui Ye, Shuo Tang, Rui Ge, Yaxin Du, Zhenfei Yin, Siheng Chen, and Jing Shao. 2025 b . MAS-GPT : Training LLMs to Build LLM-based Multi-Agent Systems . In Forty-Second International Conference on Machine Learning

    2025

  38. [38]

    Enhao Zhang, Erkang Zhu, Gagan Bansal, Adam Fourney, Hussein Mozannar, and Jack Gerrits. 2025 a . https://doi.org/10.48550/arXiv.2507.08944 Optimizing Sequential Multi-Step Tasks with Parallel LLM Agents . Preprint, arXiv:2507.08944

    work page doi:10.48550/arxiv.2507.08944 2025

  39. [39]

    Guibin Zhang, Yanwei Yue, Zhixun Li, Sukwon Yun, Guancheng Wan, Kun Wang, Dawei Cheng, Jeffrey Xu Yu, and Tianlong Chen. 2024. Cut the Crap : An Economical Communication Pipeline for LLM-based Multi-Agent Systems . In The Thirteenth International Conference on Learning Representations

    2024

  40. [40]

    Wentao Zhang, Liang Zeng, Yuzhen Xiao, Yongcong Li, Ce Cui, Yilei Zhao, Rui Hu, Yang Liu, Yahui Zhou, and Bo An. 2025 b . https://doi.org/10.48550/arXiv.2506.12508 AgentOrchestra : A Hierarchical Multi-Agent Framework for General-Purpose Task Solving . Preprint, arXiv:2506.12508

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2506.12508 2025

  41. [41]

    Andrew Zhao, Daniel Huang, Quentin Xu, Matthieu Lin, Yong-Jin Liu, and Gao Huang. 2024. https://arxiv.org/abs/2308.10144 ExpeL : LLM agents are experiential learners . In Proceedings of the AAAI Conference on Artificial Intelligence

    work page arXiv 2024

  42. [42]

    Xiaochen Zhu, Caiqi Zhang, Tom Stafford, Nigel Collier, and Andreas Vlachos. 2025. https://doi.org/10.18653/v1/2025.acl-long.195 Conformity in Large Language Models . In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics ( Volume 1: Long Papers ) , pages 3854--3872, Vienna, Austria. Association for Computational Linguistics

    work page doi:10.18653/v1/2025.acl-long.195 2025

  43. [43]

    Mingchen Zhuge, Wenyi Wang, Louis Kirsch, Francesco Faccio, Dmitrii Khizbullin, and J \"u rgen Schmidhuber. 2024. GPTSwarm : Language Agents as Optimizable Graphs . In Forty-First International Conference on Machine Learning

    2024

  44. [44]

    online" 'onlinestring :=

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

  45. [45]

    write newline

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