pith. machine review for the scientific record. sign in

arxiv: 2604.02674 · v1 · submitted 2026-04-03 · 💻 cs.MA · cs.AI

Recognition: no theorem link

Do Agent Societies Develop Intellectual Elites? The Hidden Power Laws of Collective Cognition in LLM Multi-Agent Systems

Kavana Venkatesh , Jiaming Cui
Authors on Pith no claims yet

Pith reviewed 2026-05-13 19:08 UTC · model grok-4.3

classification 💻 cs.MA cs.AI
keywords multi-agent systemsLLM coordinationpower lawspreferential attachmentintellectual elitesintegration bottleneckcollective cognitioncoordination cascades
0
0 comments X

The pith

LLM multi-agent systems develop intellectual elites as coordination follows heavy-tailed cascades and preferential attachment.

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

This paper examines coordination in societies of LLM-based agents through a large-scale empirical analysis of over 1.5 million interactions. It establishes that coordination manifests as heavy-tailed cascades which concentrate among a few agents via preferential attachment, forming intellectual elites. These patterns couple with an integration bottleneck in which expansion grows with system size while consolidation lags, leading to more frequent extreme events at larger scales. The authors introduce Deficit-Triggered Integration to selectively strengthen consolidation under imbalance and demonstrate that it improves performance exactly where coordination breaks down without restricting large-scale processes.

Core claim

The study reconstructs reasoning in LLM multi-agent systems as cascades of atomic coordination events. Analysis reveals three coupled laws: coordination cascades follow heavy-tailed distributions, participation concentrates into intellectual elites through preferential attachment, and extreme events grow more frequent with increasing system size. These are unified by an integration bottleneck where coordination expansion scales with size but consolidation does not, producing large yet weakly integrated collective reasoning. Deficit-Triggered Integration corrects the imbalance by boosting integration selectively and improves outcomes where standard coordination fails.

What carries the argument

The integration bottleneck, in which coordination expansion scales with system size while consolidation does not, which links heavy-tailed cascades, preferential attachment to elites, and rising extreme events.

If this is right

  • Coordination concentrates into intellectual elites through preferential attachment as agent numbers grow.
  • Extreme coordination events become more frequent with larger system sizes due to the mismatch in expansion and consolidation.
  • Deficit-Triggered Integration improves performance by addressing integration imbalances without suppressing large-scale reasoning.

Where Pith is reading between the lines

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

  • Designers of agent systems could monitor integration balance in real time to prevent elite concentration and maintain broader participation.
  • The power-law structure suggests that fixed topologies may require active adjustment mechanisms to sustain stable collective performance at scale.
  • Applying the same event-level cascade reconstruction to human collaboration data could reveal whether similar bottlenecks appear in non-LLM collective cognition.

Load-bearing premise

The load-bearing premise is that the atomic event-level formulation accurately reconstructs reasoning as cascades of coordination and that the observed patterns generalize beyond the specific tasks, topologies, and LLM models tested.

What would settle it

Measuring the size distribution of coordination cascades and the attachment probabilities of agents in new experiments that alter integration rules or increase agent counts would test whether the bottleneck produces the claimed laws.

Figures

Figures reproduced from arXiv: 2604.02674 by Jiaming Cui, Kavana Venkatesh.

Figure 1
Figure 1. Figure 1: Heavy-tailed coordination cascades across observables. CCDFs show a power-law regime (2 < α <ˆ 3) with truncation at large x. Dashed lines indicate MLE fits above xmin. Truncated power laws are favored over log-normal and exponential alternatives ( [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Finite-size stability of heavy-tailed coordination dynamics. (Left) Estimated tail exponents αˆ (MLE) vs. agent count N. Estimates fluctuate at small N due to limited tail samples, then stabilize and converge beyond N ≈ 64, indicating emergence of a consistent heavy-tailed regime. (Right) Mean maximum event size ⟨xmax⟩ vs. N. The upper tail grows systematically across observables, with strongest expansion … view at source ↗
Figure 3
Figure 3. Figure 3: Topology- and task-specific heavy-tailed coordination cascades in multi-agent LLM systems. Complementary cumulative distribution functions (CCDF) of coordination-event sizes P(X ≥ x) across four coordination observables: Delegation Cascade, Revision Wave, Contradiction Burst, and Total Coordination Effort (TCE) under four agent interaction topologies (Chain, Star, Hierarchical, and Dynamic Reputation) and … view at source ↗
Figure 4
Figure 4. Figure 4: provides the primary evidence for concentration. The effort shares of the top-k active agents, Eactive 10 , Eactive 25 , and Eactive 50 , lie well above their egalitarian baselines across all scales, and this gap widens systematically with N: the top-10% excess reaches +24pp at large N. The cumulative concentration curves Sp become increasingly convex, showing that larger societies develop broader and more… view at source ↗
Figure 5
Figure 5. Figure 5: Preferential attachment is a core micro-mechanism behind heavy-tailed coordination and elite concentration in LLM agent societies. (a) The routing ratio R(x, N) rises above the null baseline once a claim accumulates prior engagement, and the effect strengthens with system size N, revealing scale-dependent preferential amplification before saturating in the tail. (b) Estimated attachment slopes βˆ vary syst… view at source ↗
Figure 6
Figure 6. Figure 6: Conflict-integration dynamics and performance degradation in the high-intensity regime. (a) Mean task success across coordination intensity regimes reveals a non-monotonic signature: performance plateaus at moderate intensity before undergoing significant degradation in the high-intensity tail. This decline is strongly correlated with an elevated contradiction burden and a concomitant collapse in merge con… view at source ↗
Figure 7
Figure 7. Figure 7: Extreme-value scaling of coordination cascades. Mean maximum event size ⟨xmax⟩ vs. agent count N (log–log). Solid curves show empirical values with 95% CIs; dashed lines denote power-law fits (γˆ). All observables scale with N, with TCE showing the strongest growth and closest alignment to EVT predictions (γˆTCE ≈ 0.85 vs. γth ≈ 0.82), indicating systematic expansion of the coordination tail. Revision Dele… view at source ↗
Figure 8
Figure 8. Figure 8: Deficit-Triggered Integration (DTI) in coordination cascades. A cascade initially expands through parallel revision, contradiction, and delegation branches, leading to increasing fragmentation (left). DTI monitors this imbalance and triggers integration when it exceeds a threshold, consolidating active branch heads into a unified representation (middle). The cascade then resumes from this integrated state,… view at source ↗
Figure 9
Figure 9. Figure 9: Impact of Deficit-Triggered Integration (DTI) on collective cognition dynamics. (a) DTI preserves the heavy-tailed structure of coordination cascades while shifting truncation earlier, reducing excess tail mass without altering the intermediate scaling regime. Fixed-interval intervention, in contrast, introduces premature truncation and distorts the tail. (b) The growth of extreme coordination events with … view at source ↗
Figure 10
Figure 10. Figure 10: Heterogeneity of DTI gains across topology and task family. Relative improvement in task success (%) under DTI versus baseline, reported per topology-task condition. Gains range from +2.07% (QA × Chain) to +12.34% (Planning × Mesh/FC). DTI produces the largest improvements in conditions exhibiting the strongest expansion-integration imbalance in baseline coordination dynamics. Row and column marginals sho… view at source ↗
Figure 11
Figure 11. Figure 11: Internal composition of claim-level coordination cascades and the scale-conditioned integration bottleneck. (a) Claim-rooted cascades are grouped by total cognitive effort (TCE) quantile, pooling all tasks, topologies, and agent scales. As cascades move into the far tail, their internal event composition shifts toward delegation and contradiction, while merge remains com￾paratively weak and increasingly s… view at source ↗
Figure 12
Figure 12. Figure 12: Hierarchy of coordination structures in a multi-agent system. A task defines the global [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Event-induced transformations of the claim structure. Revision produces linear chains, [PITH_FULL_IMAGE:figures/full_fig_p027_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Delegation and cascade structure. Delegation events construct a subtask tree (left), [PITH_FULL_IMAGE:figures/full_fig_p028_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: From structured event traces to coordination cascades. Logged event fields define parent– [PITH_FULL_IMAGE:figures/full_fig_p030_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: End-to-end coordination pipeline in our experimental setup. Left: task expansion produces [PITH_FULL_IMAGE:figures/full_fig_p031_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Coordination-law flower signatures across models. Each panel summarizes one model using five global event observables: delegation cascade, revision wave, contradiction burst, merge fan-in, and total cognitive effort (TCE). Petal extent jointly reflects four law dimensions: heavier tails (lower αˆ), larger truncation scale xˆc, stronger preferential reinforcement βˆ, and greater elite concentration Eall 10… view at source ↗
Figure 18
Figure 18. Figure 18: Workload-expansion validation across agent society size. (a) Active-agent fraction A(N) for four task families remains high across scales, staying above 80% even at N = 512, indicating sustained participation. (b) Agents per subtask induced by the expansion rule, compared to the target scaling N/⌈N0.65⌉. The gradual increase from ∼2 to ∼9 agents per subtask shows that workload grows with N without over-co… view at source ↗
read the original abstract

Large Language Model (LLM) multi-agent systems are increasingly deployed as interacting agent societies, yet scaling these systems often yields diminishing or unstable returns, the causes of which remain poorly understood. We present the first large-scale empirical study of coordination dynamics in LLM-based multi-agent systems, introducing an atomic event-level formulation that reconstructs reasoning as cascades of coordination. Analyzing over 1.5 Million interactions across tasks, topologies, and scales, we uncover three coupled laws: coordination follows heavy-tailed cascades, concentrates via preferential attachment into intellectual elites, and produces increasingly frequent extreme events as system size grows. We show that these effects are coupled through a single structural mechanism: an integration bottleneck, in which coordination expansion scales with system size while consolidation does not, producing large but weakly integrated reasoning processes. To test this mechanism, we introduce Deficit-Triggered Integration (DTI), which selectively increases integration under imbalance. DTI improves performance precisely where coordination fails, without suppressing large-scale reasoning. Together, our results establish quantitative laws of collective cognition and identify coordination structure as a fundamental, previously unmeasured axis for understanding and improving scalable multi-agent intelligence.

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 / 2 minor

Summary. The paper presents the first large-scale empirical study of coordination dynamics in LLM multi-agent systems, analyzing over 1.5 million interactions across tasks, topologies, and scales. It introduces an atomic event-level formulation that reconstructs reasoning as cascades of coordination and reports three coupled laws: heavy-tailed cascades, concentration into intellectual elites via preferential attachment, and increasing frequency of extreme events with system size. These are attributed to a single structural mechanism—an integration bottleneck where coordination expansion scales with system size while consolidation does not—and tested via a Deficit-Triggered Integration (DTI) intervention that selectively boosts integration under imbalance and improves performance where coordination fails.

Significance. If the empirical patterns and mechanism hold after addressing methodological gaps, the work would establish quantitative laws of collective cognition in multi-agent LLM systems, identifying coordination structure as a previously unmeasured axis for scaling behavior and providing a targeted intervention (DTI) that preserves large-scale reasoning. The scale of the interaction dataset and the concrete DTI test represent strengths in empirical grounding.

major comments (2)
  1. [Abstract] Abstract and mechanism section: The integration bottleneck is presented as the coupling mechanism linking the three laws, but lacks an explicit quantitative derivation or falsifiable metric separating expansion (e.g., cascade size growth) from consolidation (e.g., integration depth or elite concentration) rates. Without separate scaling plots or regression coefficients for these rates, the coupling risks being inferred post-hoc from the same cascade statistics used to establish the laws.
  2. [Methods] Methods and results sections: The manuscript reports 1.5 million interactions and a concrete DTI intervention but omits detailed data exclusion rules, statistical controls, and full methods description. This makes it impossible to verify whether the observed heavy tails, elite concentration, and extreme-event frequency are driven by post-hoc choices, model-specific artifacts, or the claimed structural mechanism.
minor comments (2)
  1. Clarify notation for atomic event-level formulation and cascade reconstruction to ensure reproducibility across different LLM models and topologies.
  2. Add explicit comparisons of DTI performance against baselines in tables or figures, including effect sizes and confidence intervals.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive review and for highlighting the empirical strengths of the work. We address both major comments by adding explicit quantitative derivations and scaling analyses for the integration bottleneck, as well as a fully expanded methods section with exclusion rules, controls, and reproducibility details.

read point-by-point responses

  1. Referee: [Abstract] Abstract and mechanism section: The integration bottleneck is presented as the coupling mechanism linking the three laws, but lacks an explicit quantitative derivation or falsifiable metric separating expansion (e.g., cascade size growth) from consolidation (e.g., integration depth or elite concentration) rates. Without separate scaling plots or regression coefficients for these rates, the coupling risks being inferred post-hoc from the same cascade statistics used to establish the laws.

    Authors: We agree that an explicit quantitative separation is needed. In the revision we add Section 3.3 deriving the bottleneck: expansion is quantified as the scaling of total cascade size with system size N, while consolidation is quantified as the scaling of mean integration depth (events per elite agent) and elite concentration (Gini coefficient of participation). New Figure 4 presents separate log-log plots with fitted exponents: cascade size scales as N^1.38 (R^2=0.94), integration depth scales as log(N) (R^2=0.87), and elite concentration as N^0.21. These rates are tested against a null model of uniform random coordination; the observed divergence is statistically significant (p<0.001). The coupling is now derived from the differential scaling rather than inferred post-hoc. revision: yes

  2. Referee: [Methods] Methods and results sections: The manuscript reports 1.5 million interactions and a concrete DTI intervention but omits detailed data exclusion rules, statistical controls, and full methods description. This makes it impossible to verify whether the observed heavy tails, elite concentration, and extreme-event frequency are driven by post-hoc choices, model-specific artifacts, or the claimed structural mechanism.

    Authors: We acknowledge the gap in methodological transparency. The revised manuscript expands Section 2 with: (1) explicit exclusion rules (cascades shorter than 5 events or failing task completion are removed, comprising 1.8% of raw data); (2) statistical controls including fixed seeds, temperature fixed at 0.7, and robustness checks across GPT-4, Claude-3, and Llama-3-70B; (3) complete description of the interaction generation pipeline, topology sampling, and task distributions; (4) pre-registration note and full analysis scripts in the supplement. These additions allow independent verification that the reported laws are not artifacts of post-hoc filtering. revision: yes

Circularity Check

0 steps flagged

Empirical observations with post-hoc explanatory mechanism; no derivation reduces to inputs by construction

full rationale

The paper reports large-scale empirical measurements of coordination cascades across 1.5M interactions and proposes an integration bottleneck as a coupling explanation after the patterns are observed. No equations or fitted parameters are shown to be renamed as predictions, no self-citation chain carries the central claim, and the atomic event formulation is presented as a measurement tool rather than a self-defining loop. The DTI intervention is introduced as a test rather than a necessary consequence of the data. This is the normal case of an empirical study whose central results remain independent of any circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the assumption that event-level coordination traces faithfully represent collective reasoning and that the observed statistical patterns are not artifacts of the chosen LLMs or tasks. No free parameters are explicitly fitted to derive the laws; the integration bottleneck is postulated post-measurement.

axioms (1)
  • domain assumption Heavy-tailed distributions and preferential attachment describe coordination events in LLM agent interactions
    Invoked to interpret the measured cascades and elite formation as general structural features rather than model-specific artifacts.
invented entities (1)
  • integration bottleneck no independent evidence
    purpose: Explains why coordination expansion and consolidation do not scale together, coupling the three observed laws
    Postulated mechanism introduced after observing the patterns; no independent falsifiable prediction is provided in the abstract.

pith-pipeline@v0.9.0 · 5502 in / 1432 out tokens · 38504 ms · 2026-05-13T19:08:12.764992+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

70 extracted references · 70 canonical work pages · 9 internal anchors

  1. [1]

    Sulla determinazione empirica di una legge didistribuzione.Giorn Dell’inst Ital Degli Att, 4:89–91, 1933

    Kolmogorov An. Sulla determinazione empirica di una legge didistribuzione.Giorn Dell’inst Ital Degli Att, 4:89–91, 1933

    work page 1933

  2. [2]

    Self-organized criticality: An explanation of the 1/f noise.Physical review letters, 59(4):381, 1987

    Per Bak, Chao Tang, and Kurt Wiesenfeld. Self-organized criticality: An explanation of the 1/f noise.Physical review letters, 59(4):381, 1987

    work page 1987

  3. [3]

    Everyone’s an influencer: quantifying influence on twitter

    Eytan Bakshy, Jake M Hofman, Winter A Mason, and Duncan J Watts. Everyone’s an influencer: quantifying influence on twitter. InProceedings of the fourth ACM international conference on Web search and data mining, pages 65–74, 2011

    work page 2011

  4. [4]

    The origin of bursts and heavy tails in human dynamics.Nature, 435(7039):207–211, 2005

    Albert-Laszlo Barabasi. The origin of bursts and heavy tails in human dynamics.Nature, 435(7039):207–211, 2005

    work page 2005

  5. [5]

    Emergence of scaling in random networks.science, 286(5439):509–512, 1999

    Albert-László Barabási and Réka Albert. Emergence of scaling in random networks.science, 286(5439):509–512, 1999

    work page 1999

  6. [6]

    Why Do Multi-Agent LLM Systems Fail?

    Mert Cemri, Melissa Z Pan, Shuyi Yang, Lakshya A Agrawal, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Dan Klein, Kannan Ramchandran, et al. Why do multi- agent llm systems fail?arXiv preprint arXiv:2503.13657, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  7. [7]

    Measuring user influence in twitter: The million follower fallacy

    Meeyoung Cha, Hamed Haddadi, Fabricio Benevenuto, and Krishna Gummadi. Measuring user influence in twitter: The million follower fallacy. InProceedings of the international AAAI conference on web and social media, volume 4, pages 10–17, 2010

    work page 2010

  8. [8]

    ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate

    Chi-Min Chan, Weize Chen, Yusheng Su, Jianxuan Yu, Wei Xue, Shanghang Zhang, Jie Fu, and Zhiyuan Liu. Chateval: Towards better llm-based evaluators through multi-agent debate. arXiv preprint arXiv:2308.07201, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  9. [9]

    Langgraph: Building stateful, multi-agent applications with llms, 2024

    Harrison Chase and LangChain Inc. Langgraph: Building stateful, multi-agent applications with llms, 2024

    work page 2024

  10. [10]

    Reconcile: Round-table conference improves reasoning via consensus among diverse llms

    Justin Chen, Swarnadeep Saha, and Mohit Bansal. Reconcile: Round-table conference improves reasoning via consensus among diverse llms. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7066–7085, 2024

    work page 2024

  11. [11]

    Are more llm calls all you need? towards the scaling properties of compound ai systems

    Lingjiao Chen, Jared Davis, Boris Hanin, Peter Bailis, Ion Stoica, Matei Zaharia, and James Zou. Are more llm calls all you need? towards the scaling properties of compound ai systems. Advances in Neural Information Processing Systems, 37:45767–45790, 2024

    work page 2024

  12. [12]

    Agentverse: Facilitating multi-agent collaboration and exploring emergent behaviors

    Weize Chen, Yusheng Su, Jingwei Zuo, Cheng Yang, Chenfei Yuan, Chi-Min Chan, Heyang Yu, Yaxi Lu, Yi-Hsin Hung, Chen Qian, et al. Agentverse: Facilitating multi-agent collaboration and exploring emergent behaviors. InThe Twelfth International Conference on Learning Representations, 2023

    work page 2023

  13. [13]

    Power-law distributions in empirical data.SIAM review, 51(4):661–703, 2009

    Aaron Clauset, Cosma Rohilla Shalizi, and Mark EJ Newman. Power-law distributions in empirical data.SIAM review, 51(4):661–703, 2009

    work page 2009

  14. [14]

    Robust dynamic classes revealed by measuring the response function of a social system.Proceedings of the National Academy of Sciences, 105(41):15649– 15653, 2008

    Riley Crane and Didier Sornette. Robust dynamic classes revealed by measuring the response function of a social system.Proceedings of the National Academy of Sciences, 105(41):15649– 15653, 2008

    work page 2008

  15. [15]

    Springer, 2006

    Laurens De Haan and Ana Ferreira.Extreme value theory: an introduction. Springer, 2006

    work page 2006

  16. [16]

    Improv- ing factuality and reasoning in language models through multiagent debate

    Yilun Du, Shuang Li, Antonio Torralba, Joshua B Tenenbaum, and Igor Mordatch. Improv- ing factuality and reasoning in language models through multiagent debate. InForty-first international conference on machine learning, 2024

    work page 2024

  17. [17]

    Springer Science & Business Media, 2013

    Paul Embrechts, Claudia Klüppelberg, and Thomas Mikosch.Modelling extremal events: for insurance and finance, volume 33. Springer Science & Business Media, 2013

    work page 2013

  18. [18]

    arXiv preprint arXiv:2502.18836 , year=

    Longling Geng and Edward Y Chang. Realm-bench: A benchmark for evaluating multi- agent systems on real-world, dynamic planning and scheduling tasks.arXiv preprint arXiv:2502.18836, 2025. 17

    work page arXiv 2025

  19. [19]

    Measurement of inequality of incomes.The economic journal, 31(121):124–125, 1921

    Corrado Gini. Measurement of inequality of incomes.The economic journal, 31(121):124–125, 1921

    work page 1921

  20. [20]

    Universal behavior of load distribution in scale-free networks.Physical review letters, 87(27):278701, 2001

    K-I Goh, Byungnam Kahng, and Doochul Kim. Universal behavior of load distribution in scale-free networks.Physical review letters, 87(27):278701, 2001

    work page 2001

  21. [21]

    Problems with fitting to the power- law distribution.The European Physical Journal B-Condensed Matter and Complex Systems, 41(2):255–258, 2004

    Michel L Goldstein, Steven A Morris, and Gary G Yen. Problems with fitting to the power- law distribution.The European Physical Journal B-Condensed Matter and Complex Systems, 41(2):255–258, 2004

    work page 2004

  22. [22]

    arXiv preprint arXiv:2411.06559 , year=

    Yu Gu, Kai Zhang, Yuting Ning, Boyuan Zheng, Boyu Gou, Tianci Xue, Cheng Chang, Sanjari Srivastava, Yanan Xie, Peng Qi, et al. Is your llm secretly a world model of the internet? model-based planning for web agents.arXiv preprint arXiv:2411.06559, 2024

    work page arXiv 2024

  23. [23]

    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. Large language model based multi-agents: A survey of progress and challenges.arXiv preprint arXiv:2402.01680, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  24. [24]

    The rise and decline of an open collaboration system: How wikipedia’s reaction to popularity is causing its decline

    Aaron Halfaker, R Stuart Geiger, Jonathan T Morgan, and John Riedl. The rise and decline of an open collaboration system: How wikipedia’s reaction to popularity is causing its decline. American behavioral scientist, 57(5):664–688, 2013

    work page 2013

  25. [25]

    Metagpt: Meta programming for a multi-agent collaborative framework

    Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, et al. Metagpt: Meta programming for a multi-agent collaborative framework. InThe twelfth international conference on learning representations, 2023

    work page 2023

  26. [26]

    SWE-bench: Can Language Models Resolve Real-World GitHub Issues?

    Carlos E Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. Swe-bench: Can language models resolve real-world github issues?arXiv preprint arXiv:2310.06770, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  27. [27]

    Siegel, Nitya Nadgir, and Arvind Narayanan

    Sayash Kapoor, Benedikt Stroebl, Zachary S Siegel, Nitya Nadgir, and Arvind Narayanan. Ai agents that matter.arXiv preprint arXiv:2407.01502, 2024

    work page arXiv 2024

  28. [28]

    Towards a Science of Scaling Agent Systems

    Yubin Kim, Ken Gu, Chanwoo Park, Chunjong Park, Samuel Schmidgall, A Ali Heydari, Yao Yan, Zhihan Zhang, Yuchen Zhuang, Mark Malhotra, et al. Towards a science of scaling agent systems.arXiv preprint arXiv:2512.08296, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  29. [29]

    Highly clustered scale-free networks.Physical Review E, 65(3):036123, 2002

    Konstantin Klemm and Victor M Eguiluz. Highly clustered scale-free networks.Physical Review E, 65(3):036123, 2002

    work page 2002

  30. [30]

    Large language models miss the multi-agent mark

    Emanuele La Malfa, Gabriele La Malfa, Samuele Marro, Jie M Zhang, Elizabeth Black, Michael Luck, Philip Torr, and Michael Wooldridge. Large language models miss the multi-agent mark. arXiv preprint arXiv:2505.21298, 2025

    work page arXiv 2025

  31. [31]

    Multi- agent reinforcement learning in sequential social dilemmas.arXiv preprint arXiv:1702.03037, 2017

    Joel Z Leibo, Vinicius Zambaldi, Marc Lanctot, Janusz Marecki, and Thore Graepel. Multi- agent reinforcement learning in sequential social dilemmas.arXiv preprint arXiv:1702.03037, 2017

    work page arXiv 2017

  32. [32]

    The dynamics of viral marketing

    Jure Leskovec, Lada A Adamic, and Bernardo A Huberman. The dynamics of viral marketing. ACM Transactions on the Web (TWEB), 1(1):5–es, 2007

    work page 2007

  33. [33]

    Camel: Communicative agents for" mind" exploration of large language model society.Advances in neural information processing systems, 36:51991–52008, 2023

    Guohao Li, Hasan Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. Camel: Communicative agents for" mind" exploration of large language model society.Advances in neural information processing systems, 36:51991–52008, 2023

    work page 2023

  34. [34]

    Encouraging divergent thinking in large language models through multi- agent debate

    Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Shuming Shi, and Zhaopeng Tu. Encouraging divergent thinking in large language models through multi- agent debate. InProceedings of the 2024 conference on empirical methods in natural language processing, pages 17889–17904, 2024

    work page 2024

  35. [35]

    AgentBench: Evaluating LLMs as Agents

    Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, et al. Agentbench: Evaluating llms as agents.arXiv preprint arXiv:2308.03688, 2023. 18

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  36. [36]

    Dynamic llm-agent network: An llm-agent collaboration framework with agent team optimization

    Zijun Liu, Yanzhe Zhang, Peng Li, Yang Liu, and Diyi Yang. Dynamic llm-agent net- work: An llm-agent collaboration framework with agent team optimization.arXiv preprint arXiv:2310.02170, 2023

    work page arXiv 2023

  37. [37]

    Methods of measuring the concentration of wealth.Publications of the American statistical association, 9(70):209–219, 1905

    Max O Lorenz. Methods of measuring the concentration of wealth.Publications of the American statistical association, 9(70):209–219, 1905

    work page 1905

  38. [38]

    Multi-agent actor-critic for mixed cooperative-competitive environments.Advances in neural information processing systems, 30, 2017

    Ryan Lowe, Yi I Wu, Aviv Tamar, Jean Harb, OpenAI Pieter Abbeel, and Igor Mordatch. Multi-agent actor-critic for mixed cooperative-competitive environments.Advances in neural information processing systems, 30, 2017

    work page 2017

  39. [39]

    MIT press, 2015

    Thomas W Malone and Michael Bernstein.Handbook of collective intelligence. MIT press, 2015

    work page 2015

  40. [40]

    Gaia: a benchmark for general ai assistants

    Grégoire Mialon, Clémentine Fourrier, Thomas Wolf, Yann LeCun, and Thomas Scialom. Gaia: a benchmark for general ai assistants. InThe Twelfth International Conference on Learning Representations, 2023

    work page 2023

  41. [41]

    A brief history of generative models for power law and lognormal distributions.Internet mathematics, 1(2):226–251, 2004

    Michael Mitzenmacher. A brief history of generative models for power law and lognormal distributions.Internet mathematics, 1(2):226–251, 2004

    work page 2004

  42. [42]

    The structure and function of complex networks.SIAM review, 45(2):167– 256, 2003

    Mark EJ Newman. The structure and function of complex networks.SIAM review, 45(2):167– 256, 2003

    work page 2003

  43. [43]

    Power laws, pareto distributions and zipf’s law.Contemporary physics, 46(5):323–351, 2005

    Mark EJ Newman. Power laws, pareto distributions and zipf’s law.Contemporary physics, 46(5):323–351, 2005

    work page 2005

  44. [44]

    From text to life: On the reciprocal relationship between artificial life and large language models

    Eleni Nisioti, Claire Glanois, Elias Najarro, Andrew Dai, Elliot Meyerson, Joachim Winther Pedersen, Laetitia Teodorescu, Conor F Hayes, Shyam Sudhakaran, and Sebastian Risi. From text to life: On the reciprocal relationship between artificial life and large language models. In Artificial Life Conference Proceedings 36, volume 2024, page 39. MIT Press One...

    work page 2024

  45. [45]

    The influence of scaffolds on coordination scaling laws in LLM agents

    Rebecka Nordenlöw et al. The influence of scaffolds on coordination scaling laws in LLM agents. InNeurIPS 2025 Workshop on Multi-Turn Interactions in Large Language Models (MTI-LLM), 2025

    work page 2025

  46. [46]

    MemGPT: Towards LLMs as Operating Systems

    C Packer, V Fang, SG Patil, K Lin, S Wooders, and J Gonzalez. Memgpt: Towards llms as operating systems. arxiv 2023.arXiv preprint arXiv:2310.08560

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  47. [47]

    Librairie Droz, 1964

    Vilfredo Pareto.Cours d’économie politique, volume 1. Librairie Droz, 1964

    work page 1964

  48. [48]

    Generative agents: Interactive simulacra of human behavior

    Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. Generative agents: Interactive simulacra of human behavior. InProceed- ings of the 36th annual acm symposium on user interface software and technology, pages 1–22, 2023

    work page 2023

  49. [49]

    Harvard University Press, 2014

    Thomas Piketty.Capital in the twenty-first century. Harvard University Press, 2014

    work page 2014

  50. [50]

    Chatdev: Communicative agents for software development

    Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, et al. Chatdev: Communicative agents for software development. InProceedings of the 62nd annual meeting of the association for computational linguistics (volume 1: Long papers), pages 15174–15186, 2024

    work page 2024

  51. [51]

    Scaling large-language-model-based multi-agent collaboration

    Chen Qian, Zihao Xie, Yifei Wang, Wei Liu, Kunlun Zhu, Hanchen Xia, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, et al. Scaling large language model-based multi-agent collabora- tion.arXiv preprint arXiv:2406.07155, 2024

    work page arXiv 2024

  52. [52]

    Reflexion: Language agents with verbal reinforcement learning.Advances in neural information processing systems, 36:8634–8652, 2023

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning.Advances in neural information processing systems, 36:8634–8652, 2023

    work page 2023

  53. [53]

    Table for estimating the goodness of fit of empirical distributions.The annals of mathematical statistics, 19(2):279–281, 1948

    Nickolay Smirnov. Table for estimating the goodness of fit of empirical distributions.The annals of mathematical statistics, 19(2):279–281, 1948. 19

    work page 1948

  54. [54]

    Critical truths about power laws.Science, 335(6069):665–666, 2012

    Michael PH Stumpf and Mason A Porter. Critical truths about power laws.Science, 335(6069):665–666, 2012

    work page 2012

  55. [55]

    Vintage, 2005

    James Surowiecki.The wisdom of crowds. Vintage, 2005

    work page 2005

  56. [56]

    Multi-Agent Collaboration Mechanisms: A Survey of LLMs

    Khanh-Tung Tran, Dung Dao, Minh-Duong Nguyen, Quoc-Viet Pham, Barry O’Sullivan, and Hoang D Nguyen. Multi-agent collaboration mechanisms: A survey of llms.arXiv preprint arXiv:2501.06322, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  57. [57]

    Physicsagentabm: Physics-guided generative agent-based modeling.arXiv preprint arXiv:2602.06030, 2026

    Kavana Venkatesh, Yinhan He, Jundong Li, and Jiaming Cui. Physicsagentabm: Physics-guided generative agent-based modeling.arXiv preprint arXiv:2602.06030, 2026

    work page arXiv 2026

  58. [58]

    Power-law distributions and binned empirical data

    Yogesh S Virkar. Power-law distributions and binned empirical data. Master’s thesis, University of Colorado at Boulder, 2012

    work page 2012

  59. [59]

    Likelihood ratio tests for model selection and non-nested hypotheses.Econo- metrica: journal of the Econometric Society, pages 307–333, 1989

    Quang H Vuong. Likelihood ratio tests for model selection and non-nested hypotheses.Econo- metrica: journal of the Econometric Society, pages 307–333, 1989

    work page 1989

  60. [60]

    A survey on large language model based autonomous agents.Frontiers of Computer Science, 18(6):186345, 2024

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. A survey on large language model based autonomous agents.Frontiers of Computer Science, 18(6):186345, 2024

    work page 2024

  61. [61]

    Self-Consistency Improves Chain of Thought Reasoning in Language Models

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models.arXiv preprint arXiv:2203.11171, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  62. [62]

    A simple model of global cascades on random networks.Proceedings of the National Academy of Sciences, 99(9):5766–5771, 2002

    Duncan J Watts. A simple model of global cascades on random networks.Proceedings of the National Academy of Sciences, 99(9):5766–5771, 2002

    work page 2002

  63. [63]

    Collective dynamics of ‘small-world’networks.nature, 393(6684):440–442, 1998

    Duncan J Watts and Steven H Strogatz. Collective dynamics of ‘small-world’networks.nature, 393(6684):440–442, 1998

    work page 1998

  64. [64]

    Chain-of-thought prompting elicits reasoning in large language models

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

    work page 2022

  65. [65]

    Evidence for a collective intelligence factor in the performance of human groups

    Anita Williams Woolley, Christopher F Chabris, Alex Pentland, Nada Hashmi, and Thomas W Malone. Evidence for a collective intelligence factor in the performance of human groups. science, 330(6004):686–688, 2010

    work page 2010

  66. [66]

    Autogen: Enabling next-gen llm applications via multi-agent conversations

    Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, et al. Autogen: Enabling next-gen llm applications via multi-agent conversations. InFirst conference on language modeling, 2024

    work page 2024

  67. [67]

    The rise and potential of large language model based agents: A survey.Science China Information Sciences, 68(2):121101, 2025

    Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, Yiwen Ding, Boyang Hong, Ming Zhang, Junzhe Wang, Senjie Jin, Enyu Zhou, et al. The rise and potential of large language model based agents: A survey.Science China Information Sciences, 68(2):121101, 2025

    work page 2025

  68. [68]

    React: Synergizing reasoning and acting in language models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models. InThe eleventh international conference on learning representations, 2022

    work page 2022

  69. [69]

    Multiagentbench: Evaluating the collaboration and competition of llm agents

    Kunlun Zhu, Hongyi Du, Zhaochen Hong, Xiaocheng Yang, Shuyi Guo, Daisy Zhe Wang, Zhenhailong Wang, Cheng Qian, Robert Tang, Heng Ji, et al. Multiagentbench: Evaluating the collaboration and competition of llm agents. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 8580–8622, 2025

    work page 2025

  70. [70]

    Gptswarm: Language agents as optimizable graphs

    Mingchen Zhuge, Wenyi Wang, Louis Kirsch, Francesco Faccio, Dmitrii Khizbullin, and Jürgen Schmidhuber. Gptswarm: Language agents as optimizable graphs. InForty-first International Conference on Machine Learning, 2024. 20 Appendix A Additional Qualitative Results In this section, we provide additional qualitative results pertaining to each of the hypothes...

    work page 2024