Pith. sign in

REVIEW 4 major objections 6 minor 87 references

Scalable, Symbiotic, AI and Non-AI Agent Based Parallel Discrete Event Simulations

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Orchestrating small language models with rule-based verifiers inside a discrete event simulation lifts accuracy from under 23% to 68% on four algorithmic task families.

desk verdict A plausible engineering recipe for orchestrating SLMs with rule-based verifiers on a PDES engine, but the headline accuracy claim is not backed by the reported experiments. read the letter →

arxiv 2505.23846 v1 pith:6T72XSN2 submitted 2025-05-28 cs.CL cs.MA

Atanu Barai , Stephan Eidenbenz , Nandakishore Santhi This is my paper
classification cs.CLcs.MA
keywords multi-agentsystemsparalleldiscreteeventsimulationsmalllanguagemodelsAIverificationproblemdecompositionrule-basedauditingtrustworthyalgorithmicreasoning
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

This paper claims that coupling several small language models with rule-based verifier agents inside a parallel discrete event simulation (PDES) solves algorithmic tasks those models cannot solve on their own. The authors break each task into small subtasks, let separate SLM agents answer each subtask in parallel, and have non-AI agents audit and correct the answers before the next event is scheduled. Across four domains—geometry (agent gathering), combinatorics (sorting), arithmetic (long multiplication), and graph theory (breadth-first traversal)—the coupled system reaches 68% overall accuracy while the same models prompted zero-shot reach under 23%. The importance, if true, is that modest models on modest hardware can be made reliable at structured reasoning by enforcing causality, constraints, and verification at every step, rather than by scaling up model size.

What carries the argument

The load-bearing object is the PDES event loop with two kinds of entities: AI agents, each running a small language model, and non-AI agents, running deterministic checks and side calculations. An agent schedules work on another entity through a request service that inserts an event into that entity's timestamped event list; the next step in the computation happens only after the preceding event has been processed. The non-AI agent acts as an unbiased auditor at each step, verifying constraints such as the maximum-speed bound, checking the smallest-element choice, correcting partial products, or tracking visited nodes, and only then scheduling the next event. Around this loop, the paper adds a second constraint mechanism: subtasks are posed with a finite set of choices, shrinking the SLM's solution space before it answers.

What would settle it

Re-run the same ten-instance suites for sorting, multiplication, and BFS with the non-AI verification and correction events disabled while keeping the subtask decomposition and multiple-choice prompts; if the accuracy stays near the claimed 68% instead of falling toward the reported vanilla 20–26%, the auditors are not doing the work. A second check is to count, from the simulation logs, how many SLM responses lacked a parseable formatted line or parsed to a coordinate outside the stated bounds.

Watch

Extended reading notes

Core claim

The paper's central claim is that treating AI decisions as events in a parallel discrete event simulation, with non-AI agents as auditors between events, converts unreliable small-model outputs into reliable solutions of multi-step algorithmic problems. For ten randomly generated instances in each of the sorting, multiplication, and graph-traversal domains, the PDES-coupled pipeline achieves 77% accuracy with Llama-3.1 and 60% with Mistral-Nemo, versus 26% and 20% with vanilla zero-shot prompting; averaged over the reported domains and models, the method attains 68% against under 23% for vanilla models. The geometry experiments show the verifier suppressing the agents' overspeed jumps, which are visible when the SLM alone chooses positions. The authors formulate the contribution as a general framework: decompose a problem into choice-constrained subtasks, deploy many SLM entities in parallel, verify each response with deterministic rules, and let the simulation's event ordering enforce causal consistency.

Load-bearing premise

Every SLM output must arrive in a rigidly formatted, machine-parsable line (for example, New_Position:(x,y) as the last line), and the non-AI auditor must parse and correct it without error; if a model breaks the format or the parser misreads the coordinates, the simulation silently proceeds with wrong data.

Editorial extensions

If this is right

  • The same SLMs that score under 23% zero-shot on these tasks can be orchestrated to 68% accuracy, so raw model capability is not the only lever on structured reasoning.
  • Every intermediate answer is checked by a deterministic agent, so errors are caught at the step where they occur instead of propagating to a final answer.
  • Because agents are independent PDES entities, adding more SLM entities for parallel subtasks scales with available MPI ranks up to the number of agents, with simulation time decreasing and tokens generated per minute increasing.
  • The method applies across at least four distinct domains with a single generic framework, so problem-specific tuning is limited to defining subtasks and audit rules.
  • Deploying many small models instead of one large one lowers the memory and GPU requirements for strong algorithmic performance.

Reading between the lines

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

  • A testable extension is to apply the same event-driven audit pattern to tasks with softer constraints, such as code generation with a compiler as the non-AI auditor; the accuracy gain should track how completely the constraint can be checked automatically.
  • The reported accuracy gap suggests that the ceiling of the method is set by the verifier's coverage and the SLM's format compliance, not by the SLM's mathematical ability; if so, improving parser robustness should raise accuracy further.
  • Because the paper's geometry results show the auditor correcting overspeed jumps, the framework could double as a measurement tool: comparing audited versus unaudited trajectories quantifies how often a model would violate a hard constraint in a given task.
  • The approach implies a division of labor that the authors leave implicit: SLMs provide broad heuristic search over candidate choices, while deterministic agents provide the deductive steps, so the method should work best on problems where every step has a checkable invariant.
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

4 major / 6 minor

Summary. The manuscript proposes a parallel discrete event simulation (PDES) framework, built on the Simian engine, for coupling multiple small language model (SLM) agents with rule-based non-AI agents. The method decomposes a problem into subtasks, schedules them as events among SLM agents, and uses non-AI agents to verify and correct outputs and to compute structured intermediate results. The authors demonstrate the approach on four problems—gathering (geometry), sorting (combinatorics), long multiplication (arithmetic), and breadth-first traversal (graph theory)—and report an overall accuracy of 68% versus less than 23% for zero-shot SLM baselines, along with scaling experiments on up to six MPI ranks.

Significance. If the reported accuracy gains were fully substantiated, the paper would make a useful contribution: it shows that small open-weight models, embedded in a causal event-driven architecture with rule-based verifiers and multiple-choice constraints, can outperform their unassisted zero-shot performance on several structured reasoning tasks, while also benefiting from PDES parallelism. The main strengths are the concrete demonstration on four domains, the use of an established PDES engine, and the initial scaling measurements in Figures 9 and 10. However, the evaluation as presented is too incomplete to support the headline claim: the aggregate accuracy number excludes one of the four domains, lacks statistical detail, and the comparison is confounded by the fact that the non-AI agents themselves perform substantial parts of the computation. The manuscript therefore requires a major revision before its central claim can be assessed.

major comments (4)
  1. [§5.3, Fig. 8] The abstract's central claim that 'overall accuracy of our approach is 68% where as the accuracy of vanilla models is less than 23%' is not supported by the reported experiments. Figure 8 compares only sorting, multiplication, and BFS; the geometry problem is excluded from the accuracy comparison and never receives a quantitative accuracy value. The 68% figure is simply the average of the two model-level numbers given in the text (77% for Llama-3.1 and 60% for Mistral-Nemo), and the corresponding zero-shot average is exactly 23% (26% and 20%), not 'less than 23%.' No per-task accuracies, trial counts, confidence intervals, or significance tests are provided, so the aggregate could be driven by one task or by sampling noise. Please report the full per-task breakdown with N and error bars, include the geometry domain in the accuracy evaluation or explicitly state that the headline refers to three domains, and correct the abstract accordingly.
  2. [Fig. 2, lines 27-28; §4.3.1] The only code artifact, the gathering problem snippet in Figure 2, contains a bug that prevents the reported geometry results from being reproduced or trusted: lines 27 and 28 both assign the parsed coordinate to self.position[0], so the y-coordinate is overwritten by the x-coordinate. The full code is not released, and the paper does not report how often SLM responses violate the required 'New_Position:(..,..)' format or are misparsed. Since both the geometry experiment and the overall method rely on robust parsing of the last output line, please release the code, fix the snippet (or correct it if the actual implementation differs), and report the parse-failure rate and its effect on the results.
  3. [§4.3.1, §4.3.3, §4.2] The accuracy comparison is partly by construction, because the non-AI agents do not merely verify LLM output; they compute or correct the core of the answer. In the gathering problem the non-AI agent corrects speed-limit violations and computes the geometric median (§4.3.1); in multiplication the non-AI agent performs the final addition of partial products (§4.3.3); in sorting and BFS the verifier knows the ground truth. Consequently, the measured 'PDES accuracy' reflects the combined system of SLM plus rule-based computation, not the ability of the SLM agents to solve the task. The paper should be reframed accordingly and should report an ablation or a separate analysis of which subtasks are actually solved by the SLM versus the non-AI agent, or compare against a baseline in which the same non-AI post-processing is applied to the vanilla model's output.
  4. [§5.2] The choice of model per task appears to be made after observing performance: 'We use Qwen in the geometry problem, since it can provide new positions of the agents quite accurately compared to other two models' (§5.2). With no pre-registered selection rule and no error bars around the model-level accuracies, this post hoc selection risks inflating the reported gains. Please state the model-selection protocol, report results for all models on all tasks, or justify why the reported subset is representative.
minor comments (6)
  1. [Table 2] The caption is copied from Table 1 and reads 'CPU, GPU and Memory Configuration of a compute node', but the table lists model names, parameter sizes, and quantized model sizes; please correct the caption to describe the actual contents.
  2. [Figure 7] The subfigure labels are confusing: (a)-(e) are described as without verification and (f)-(j) as with verification, but the caption lists only temperature values for both groups; please add explicit 'without verification' and 'with verification' labels to the caption and the figure panels.
  3. [Section 4.4] The text states that the framework will be open-sourced, but no repository link or availability statement is provided; an artifact link would help readers verify the claims.
  4. [Figure 2, lines 40-52] The snippet contains several typos that make it hard to parse, including 'reached _optimals', 'ch oose_ne xt_step', 'rea ched_opt imals', and 'opt imal_pos ition'; these should be cleaned up if the code is meant to be illustrative.
  5. [Figures 9-10] The strong-scaling claims are reported without repeated trials or error bars, so the text should state these are single-run observations rather than statistically validated scaling results.
  6. [References] Several references contain typos, e.g., 'Septerber' in [5], [31], and [32], and the model name is written inconsistently as 'Mistral-Nemo' and 'mistral-nemo' in Figure 8; please standardize these.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the accuracy claim is an empirical evaluation of the full PDES pipeline, not a derivation from fitted inputs or a self-citation chain.

full rationale

The paper's central claim is an empirical evaluation of a proposed PDES-based pipeline, not a derivation in which an output is constructed from inputs that already contain it. The non-AI agents' verification and computation roles are explicitly part of the method under test: the Figure 2 caption states "the non-AI agent mainly performs structured mathematical side-calculations that SLMs are not so good at natively," and Section 4.3.3(d) states "The partials results from each SLM are finally added using a non-AI agent." Because the method is defined to include rule-based auditing and arithmetic, the reported accuracy gain over zero-shot prompting is a measure of that entire pipeline; it is not a hidden re-introduction of the evaluation target as an input. No parameter is fitted to the accuracy data, no prediction is derived from a fitted value, and the only self-citations (Simian engine [56] and SimX [66]) are cited as open-source infrastructure tools rather than as the source of the accuracy result. The paper does have reporting weaknesses -- the 68% headline is the average of 77% and 60% over only the sorting, multiplication, and BFS tasks, geometry is excluded from the quantitative comparison, and no error bars are given -- but these are evidentiary and correctness concerns, not circularity of the derivation chain.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No numeric parameters are fitted to data in the usual sense; the experimental settings such as temperature are choices, not fitted constants. The central claim rests on tooling and behavioral assumptions about the Simian engine and about SLM output formatting, rather than on parameterized models. No new physical or conceptual entities are introduced.

assumptions (3)
  • domain assumption Simian PDES engine correctly schedules events across MPI ranks without causality errors in the tested configurations.
    Methodology and scaling results (Sections 3.1, 4.1, 5.4) assume the engine's parallel event handling works as documented.
  • domain assumption Each SLM produces output in the exact requested format that the parser can extract, such as 'New_Position:(..,..)'.
    Sections 4.1 and 4.3 depend on this; parsing failure rates are not reported and the snippet has a parsing bug.
  • domain assumption Quantized 7B-12B instruct models are representative of small language models and behave appropriately for these tasks.
    Section 5.2 chooses these models as SLMs; the claim that SLMs cannot solve the tasks is based on this set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scalable, Symbiotic, AI and Non-AI Agent Based Parallel Discrete Event Simulations." pith.science (2026). https://pith.science/paper/6T72XSN2

@misc{pith2026250523846,
  author       = {Pith},
  title        = {Pith review of: Scalable, Symbiotic, AI and Non-AI Agent Based Parallel Discrete Event Simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6T72XSN2}},
  note         = {Machine review of arXiv:2505.23846}
}
read the original abstract

To fully leverage the potential of artificial intelligence (AI) systems in a trustworthy manner, it is desirable to couple multiple AI and non-AI systems together seamlessly for constraining and ensuring correctness of the output. This paper introduces a novel parallel discrete event simulation (PDES) based methodology to combine multiple AI and non-AI agents in a causal, rule-based way. Our approach tightly integrates the concept of passage of time, with each agent considered as an entity in the PDES framework and responding to prior requests from other agents. Such coupling mechanism enables the agents to work in a co-operative environment towards a common goal while many tasks run in parallel throughout the simulation. It further enables setting up boundaries to the outputs of the AI agents by applying necessary dynamic constraints using non-AI agents while allowing for scalability through deployment of hundreds of such agents in a larger compute cluster. Distributing smaller AI agents can enable extremely scalable simulations in the future, addressing local memory bottlenecks for model parameter storage. Within a PDES involving both AI and non-AI agents, we break down the problem at hand into structured steps, when necessary, providing a set of multiple choices to the AI agents, and then progressively solve these steps towards a final goal. At each step, the non-AI agents act as unbiased auditors, verifying each action by the AI agents so that certain rules of engagement are followed. We evaluate our approach by solving four problems from four different domains and comparing the results with those from AI models alone. Our results show greater accuracy in solving problems from various domains where the AI models struggle to solve the problems solely by themselves. Results show that overall accuracy of our approach is 68% where as the accuracy of vanilla models is less than 23%.

Figures

Figures reproduced from arXiv: 2505.23846 by the authors.

Figure 1
Figure 1. AI / non-AI agents are coupled using the Simian [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Code snippet from our simulator to solve the gathering problem involving multiple AI agent entities and a non-AI [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Simulation process for sorting array using SLM. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (7 more)
Figure 3
Figure 3. Figure 3: Simulation process of multiple agents trying to [PITH_FULL_IMAGE:figures/full_fig_p006_3.png]
Figure 5
Figure 5. Figure 5: Simulation process for long multiplication involving [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Simulation process for breadth-first traversal involv [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Comparing the paths the SLM-based agents take [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Comparison of results among PDES-based approach [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Simulation time as number of MPI ranks in the PDES [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Tokens/sec with scaling of MPI ranks user and system prompts to get the responses from SLMs for the sorting problem. Sys tem Promp t : You a r e an AI a g e n t who can s o r t an a r r a y . U se r Promp t : S o r t t h e a r r a y [ 4 7 5 6 2 3 , 5 7 7 9 6 3 , 1 5 6…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

87 extracted references · 46 canonical work pages

  1. [1]

    Eduardo Alonso, Mark D’Inverno, Daniel Kudenko, Michael Luck, and Jason Noble

  2. [2]

    July 2024

    Apple. July 2024. Metal Overview. https://developer.apple.com/metal/

    2024

  3. [3]

    Alessio Baratta, Antonio Cimino, Maria Grazia Gnoni, and Francesco Longo. 2023. Human Robot Collaboration in Industry 4.0: a literature review.Procedia Computer Science 217 (2023), 1887–1895. doi:10.1016/j.procs.2022.12.389 4th International Conference on Industry 4.0 and Smart Manufacturing

    work page doi:10.1016/j.procs.2022.12.389 2023

  4. [4]

    Barnes, Christopher D

    Peter D. Barnes, Christopher D. Carothers, David R. Jefferson, and Justin M. LaPre

  5. [5]

    Septerber 2024

    A Bartowski. Septerber 2024. Mistral-Nemo-Instruct-2407-GGUF - Hugging Face . https://huggingface.co/bartowski/Mistral-Nemo-Instruct-2407-GGUF

    2024

  6. [6]

    Regina Barzilay and Lillian Lee. 2004. Catching the Drift: Probabilistic Con- tent Models, with Applications to Generation and Summarization. In Proceed- ings of the Human Language Technology Conference of the North American Chap- ter of the Association for Computational Linguistics: HLT-NAACL 2004 . Asso- ciation for Computational Linguistics, Boston, Ma...

    2004

  7. [7]

    Vladimir Batagelj and Ulrik Brandes. 2005. Efficient generation of large random networks. Phys. Rev. E 71 (Mar 2005), 036113. Issue 3. doi:10.1103/PhysRevE.71. 036113

    doi:10.1103/physreve.71 2005

  8. [8]

    Bauer Jr., Christopher D

    David W. Bauer Jr., Christopher D. Carothers, and Akintayo Holder. 2009. Scalable Time Warp on Blue Gene Supercomputers. In2009 ACM/IEEE/SCS 23rd Workshop on Principles of Advanced and Distributed Simulation. 35–44. doi:10.1109/PADS.2009.21

    work page doi:10.1109/pads.2009.21 2009

Show all 87 references
  1. [9]

    Bengio, P

    Y. Bengio, P. Simard, and P. Frasconi. 1994. Learning long-term dependencies with gradient descent is difficult. IEEE Transactions on Neural Networks 5, 2 (1994), 157–166. doi:10.1109/72.279181

    1994 doi

  2. [10]

    Adam Berger and John Lafferty. 2001. Statistical machine learning for information retrieval. Ph. D. Dissertation. Carnegie Mellon University, USA. AAI3168516

    2001

  3. [11]

    October 2024

    Andrei Betlen. October 2024. Python bindings for llama.cpp . https://github.com/ abetlen/llama-cpp-python

    2024

  4. [12]

    Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, and Michael S

    Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, and Michael S. Bernstein et. al. 2021. On the Opportunities and Risks of Foundation Models. ArXiv (2021). https://crfm.stanford.edu/assets/report. pdf

    2021

  5. [13]

    Brown, Peter V

    Peter F. Brown, Peter V. deSouza, Robert L. Mercer, Vincent J. Della Pietra, and Jenifer C. Lai. 1992. Class-based n-gram models of natural language. Comput. Linguist. 18, 4 (Dec. 1992), 467–479

    1992

  6. [14]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Pra- fulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey...

    2020

  7. [15]

    Yu, Qiang Yang, and Xing Xie

    Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, Wei Ye, Yue Zhang, Yi Chang, Philip S. Yu, Qiang Yang, and Xing Xie. 2024. A Survey on Evaluation of Large Language Models. ACM Trans. Intell. Syst. Technol...

    2024 doi

  8. [16]

    Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W. Cohen. 2023. Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks. Transactions on Machine Learning Research (2023). https: //openreview.net/forum?id=YfZ4ZPt8zd

    2023

  9. [17]

    Kyunghyun Cho, Bart van Merriënboer, Dzmitry Bahdanau, and Yoshua Bengio

  10. [18]

    Alexis Conneau, Douwe Kiela, Holger Schwenk, Loïc Barrault, and Antoine Bordes

  11. [19]

    June 2017

    NVIDIA Corporation. June 2017. Volta Tesla V100 GPU Architecture Whitepa- per. http://images.nvidia.com/content/volta-architecture/pdf/volta-architecture- whitepaper.pdf

    2017

  12. [20]

    Bruce Croft and John Lafferty. 2003. Language modeling for information retrieval . Vol. 13. Springer Science & Business Media

    2003

  13. [21]

    Damai Dai, Chengqi Deng, Chenggang Zhao, R. X. Xu, Huazuo Gao, Deli Chen, Jiashi Li, Wangding Zeng, Xingkai Yu, Y. Wu, Zhenda Xie, Y. K. Li, Panpan Huang, Fuli Luo, Chong Ruan, Zhifang Sui, and Wenfeng Liang. 2024. DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture...

    2024 arXiv

  14. [22]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv:1810.04805 [cs.CL] https://arxiv.org/abs/1810.04805

    2019 arXiv

  15. [23]

    Georgios Diamatopoulos, Georgios Theodoropoulos, Nikos Tziritas, and Rami Bahsoon. 2024. Towards LLM Augmented Discrete Event Simulation of Blockchain Systems. In Proceedings of the 38th ACM SIGSIM Conference on Principles of Ad- vanced Discrete Simulation (Atlanta, GA, USA) (...

    2024

  16. [24]

    Ruomeng Ding, Chaoyun Zhang, Lu Wang, Yong Xu, Minghua Ma, Wei Zhang, Si Qin, Saravan Rajmohan, Qingwei Lin, and Dongmei Zhang. 2024. Everything of Thoughts: Defying the Law of Penrose Triangle for Thought Generation. In ACL (Findings). 1638–1662. https://doi.org/10.18653/v1/2...

    2024 doi

  17. [25]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, and Angela Fan et. al. 2024. The Llama 3 Herd of Models. arXiv:2407.21783 [cs.AI] https://arxiv. org/abs/2407.21783

    2024 arXiv

  18. [26]

    Badawy, Nandakishore Santhi, Stephan Eidenbenz, and Dmitry Ponomarev

    Ali Eker, Yehia Arafa, Abdel-Hameed A. Badawy, Nandakishore Santhi, Stephan Eidenbenz, and Dmitry Ponomarev. 2021. Load-Aware Dynamic Time Syn- chronization in Parallel Discrete Event Simulation. In Proceedings of the 2021 ACM SIGSIM Conference on Principles of Advanced Discre...

    2021

  19. [27]

    Bahare Fatemi, Jonathan Halcrow, and Bryan Perozzi. 2024. Talk like a Graph: En- coding Graphs for Large Language Models. InThe Twelfth International Conference on Learning Representations. https://openreview.net/forum?id=IuXR1CCrSi

    2024

  20. [28]

    Fujimoto

    Richard M. Fujimoto. 1990. Parallel discrete event simulation. Commun. ACM 33, 10 (Oct. 1990), 30–53. doi:10.1145/84537.84545

    1990

  21. [29]

    Fujimoto, K

    Richard M. Fujimoto, K. Perumalla, A. Park, H. Wu, M.H. Ammar, and G.F. Riley

  22. [30]

    Jianfeng Gao and Chin-Yew Lin. 2004. Introduction to the special issue on statistical language modeling. ACM Transactions on Asian Language Information Processing 3, 2 (June 2004), 87–93. doi:10.1145/1034780.1034781

    2004

  23. [31]

    Septerber 2024

    Georgi Gerganov. Septerber 2024. GGUF. https://github.com/ggerganov/ggml/ blob/master/docs/gguf.md

    2024

  24. [32]

    Septerber 2024

    Georgi Gerganov. Septerber 2024. LLM inference in C/C++ . https://github.com/ ggerganov/llama.cpp

    2024

  25. [33]

    Zelalem Gero, Chandan Singh, Hao Cheng, Tristan Naumann, Michel Galley, Jianfeng Gao, and Hoifung Poon. 2023. Self-Verification Improves Few-Shot Clinical Information Extraction. arXiv:2306.00024 [cs.CL] https://arxiv.org/abs/ 2306.00024

    2023 arXiv

  26. [34]

    Giabbanelli

    Philippe J. Giabbanelli. 2023. GPT-Based Models Meet Simulation: How to Effi- ciently use Large-Scale Pre-Trained Language Models Across Simulation Tasks. In Proceedings of the Winter Simulation Conference (San Antonio, Texas, USA) (WSC ’23). IEEE Press, 2920–2931

    2023

  27. [35]

    Shibo Hao, Yi Gu, Haodi Ma, Joshua Jiahua Hong, Zhen Wang, Daisy Zhe Wang, and Zhiting Hu. 2023. Reasoning with Language Model is Planning with World Model. arXiv:2305.14992 [cs.CL] https://arxiv.org/abs/2305.14992

    2023 arXiv

  28. [36]

    Pablo Hernandez-Leal, Bilal Kartal, and Matthew E Taylor. 2019. A survey and critique of multiagent deep reinforcement learning. Autonomous Agents and Multi-Agent Systems 33, 6 (2019), 750–797

    2019

  29. [37]

    Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long Short-Term Memory.Neural Computation 9, 8 (1997), 1735–1780. doi:10.1162/neco.1997.9.8.1735

    1997 doi

  30. [38]

    Takaaki Hori, Jaejin Cho, and Shinji Watanabe. 2018. End-to-end Speech Recog- nition With Word-Based Rnn Language Models. In 2018 IEEE Spoken Language Technology Workshop (SLT). 389–396. doi:10.1109/SLT.2018.8639693

    2018

  31. [39]

    Anthony Hunter, Simon Parsons, and Michael Wooldridge. 2014. Measuring inconsistency in multi-agent systems. KI-Künstliche Intelligenz 28 (2014), 169– 178

    2014

  32. [40]

    Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne...

    2024 arXiv

  33. [41]

    Tushar Khot, Harsh Trivedi, Matthew Finlayson, Yao Fu, Kyle Richardson, Peter Clark, and Ashish Sabharwal. 2023. Decomposed Prompting: A Modular Approach for Solving Complex Tasks. arXiv:2210.02406 [cs.CL] https://arxiv.org/abs/2210. 02406

    2023 arXiv

  34. [42]

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2024. Large language models are zero-shot reasoners. In Proceedings of the 36th International Conference on Neural Information Processing Systems (New Orleans, LA, USA) (NIPS ’22). Curran Associ...

    2024

  35. [43]

    Stefan Kombrink, Tomas Mikolov, Martin Karafiát, and Lukás Burget. 2011. Re- current Neural Network Based Language Modeling in Meeting Recognition.. In Interspeech, Vol. 11. 2877–2880

    2011

  36. [44]

    Alisa Liu, Xiaochuang Han, Yizhong Wang, Yulia Tsvetkov, Yejin Choi, and Noah A. Smith. 2024. Tuning Language Models by Proxy. arXiv:2401.08565 [cs.CL] https: //arxiv.org/abs/2401.08565

    2024 arXiv

  37. [45]

    Tianming Liu, Jirong Yang, and Yafeng Yin. 2024. Toward LLM-Agent-Based Model- ing of Transportation Systems: A Conceptual Framework. arXiv:2412.06681 [cs.AI] https://arxiv.org/abs/2412.06681

    2024 arXiv

  38. [46]

    Xiaoyong Liu and W Bruce Croft. 2005. Statistical language modeling for informa- tion retrieval. Annu. Rev. Inf. Sci. Technol. 39, 1 (2005), 1–31

    2005

  39. [47]

    Tomas Mikolov, Martin Karafiát, Lukas Burget, Jan Cernock`y, and Sanjeev Khu- danpur. 2010. Recurrent neural network based language model.. In Interspeech, Vol. 2. Makuhari, 1045–1048

    2010

  40. [48]

    Guo Ming. 2023. Exploration of the intelligent control system of autonomous vehicles based on edge computing. PLoS One 18, 2 (2023), e0281294

    2023

  41. [49]

    Mohammed Muqeeth, Haokun Liu, Yufan Liu, and Colin Raffel. 2024. Learn- ing to Route Among Specialized Experts for Zero-Shot Generalization. arXiv:2402.05859 [cs.LG] https://arxiv.org/abs/2402.05859

    2024 arXiv

  42. [50]

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, and Sam Altman et. al. 2024. GPT-4 Technical Report. arXiv:2303.08774 [cs.CL] https://arxiv.org/abs/ 2303.08774

    2024 arXiv

  43. [51]

    Perumalla

    Kalyan S. Perumalla. 2007. Scaling time warp-based discrete event execution to 104 processors on a Blue Gene supercomputer. In Proceedings of the 4th Interna- tional Conference on Computing Frontiers (Ischia, Italy) (CF ’07). Association for Computing Machinery, New York, NY, ...

    2007

  44. [52]

    Zhenting Qi, Mingyuan Ma, Jiahang Xu, Li Lyna Zhang, Fan Yang, and Mao Yang. 2024. Mutual Reasoning Makes Smaller LLMs Stronger Problem-Solvers. arXiv:2408.06195 [cs.CL] https://arxiv.org/abs/2408.06195

    2024 arXiv

  45. [53]

    Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, ...

    2025 arXiv

  46. [54]

    Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. 2018. Improv- ing language understanding by generative pre-training. https://cdn.openai.com/ research-covers/language-unsupervised/language_understanding_paper.pdf

    2018

  47. [55]

    Rong Rong, Jiang Hao, and Jason Liu. 2014. Performance Study of a Minimalistic Simulator on XSEDE Massively Parallel Systems. In Proceedings of the 2014 Annual Conference on Extreme Science and Engineering Discovery Environment (Atlanta, GA, USA) (XSEDE ’14). Association for C...

    2014

  48. [56]

    Nandakishore Santhi, Stephan Eidenbenz, and Jason Liu. 2015. The Simian concept: Parallel Discrete Event Simulation with interpreted languages and just-in-time compilation. In 2015 Winter Simulation Conference (WSC) . 3013–3024. doi:10.1109/ WSC.2015.7408405

    2015

  49. [57]

    Ankit Satpute, Noah Gießing, André Greiner-Petter, Moritz Schubotz, Olaf Teschke, Akiko Aizawa, and Bela Gipp. 2024. Can LLMs Master Math? Investigating Large Language Models on Math Stack Exchange. In Proceedings of the 47th Interna- tional ACM SIGIR Conference on Research an...

    2024

  50. [58]

    Noam Shazeer, *Azalia Mirhoseini, *Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. 2017. Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer. In International Conference on Learning Representations. https://openreview.net/for...

    2017

  51. [59]

    Shannon Zejiang Shen, Hunter Lang, Bailin Wang, Yoon Kim, and David Son- tag. 2024. Learning to Decode Collaboratively with Multiple Language Models. arXiv:2403.03870 [cs.CL] https://arxiv.org/abs/2403.03870

    2024 arXiv

  52. [60]

    Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language Agents with Verbal Re- inforcement Learning. arXiv:2303.11366 [cs.AI] https://arxiv.org/abs/2303.11366

    2023 arXiv

  53. [61]

    July 16, 2024

    Viraj Singh. July 16, 2024. Why LLMs Are Bad at Math — and How They Can Be Better. https://www.reachcapital.com/2024/07/16/why-llms-are-bad-at-math- and-how-they-can-be-better/

    2024

  54. [62]

    Peter Stone and Manuela Veloso. 2000. Multiagent systems: A survey from a machine learning perspective. Autonomous Robots 8 (2000), 345–383

    2000

  55. [63]

    Foteini Strati, Paul Elvinger, Tolga Kerimoglu, and Ana Klimovic. 2024. ML Train- ing with Cloud GPU Shortages: Is Cross-Region the Answer?. In Proceedings of the 4th Workshop on Machine Learning and Systems (Athens, Greece) (EuroML- Sys ’24). Association for Computing Machine...

    2024

  56. [64]

    July 2024

    Mistral AI team. July 2024. Mistral Nemo. https://mistral.ai/news/mistral-nemo/

    2024

  57. [65]

    June 2017

    TechPowerUp. June 2017. NVIDIA Tesla V100 Specs. https://www.techpowerup. com/gpu-specs/tesla-v100-pcie-16-gb.c2957

    2017

  58. [66]

    Sunil Thulasidasan, Lukas Kroc, and Stephan Eidenbenz. 2014. Developing parallel, discrete event simulations in Python - first results and user experiences with the SimX library. In 2014 4th International Conference On Simulation And Modeling Methodologies, Technologies And Ap...

    2014

  59. [67]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In Advances in Neural Information Processing Systems , I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. V...

    2017

  60. [68]

    Yu Wang, Zhiwei Liu, Jianguo Zhang, Weiran Yao, Shelby Heinecke, and Philip S. Yu. 2023. DRDT: Dynamic Reflection with Divergent Thinking for LLM-based Sequential Recommendation. arXiv:2312.11336 [cs.IR] https://arxiv.org/abs/2312. 11336

    2023 arXiv

  61. [69]

    Dai, and Quoc V Le

    Jason Wei, Maarten Bosma, Vincent Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V Le. 2022. Finetuned Language Models are Zero-Shot Learners. In International Conference on Learning Representations . https://openreview.net/forum?id=gEZrGCozdqR

    2022

  62. [70]

    Chi, Tat- sunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus

    Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tat- sunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. 2022. Emergent Abilities of Large Language M...

    2022

  63. [71]

    Chi, Quoc V

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2024. Chain-of-thought prompting elicits reasoning in large language models. In Proceedings of the 36th International Conference on Neural Information Processi...

    2024

  64. [72]

    Michael Wooldridge. 2009. An introduction to multiagent systems . John wiley & sons

    2009

  65. [73]

    Zengqing Wu, Run Peng, Xu Han, Shuyuan Zheng, Yixin Zhang, and Chuan Xiao

  66. [74]

    Fuzhao Xue, Zian Zheng, Yao Fu, Jinjie Ni, Zangwei Zheng, Wangchunshu Zhou, and Yang You. 2024. OpenMoE: An Early Effort on Open Mixture-of-Experts Language Models. arXiv:2402.01739 [cs.CL] https://arxiv.org/abs/2402.01739

    2024 arXiv

  67. [75]

    Griffiths, Yuan Cao, and Karthik Narasimhan

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. Tree of Thoughts: Deliberate Problem Solving with Large Language Models. arXiv:2305.10601 [cs.CL] https://arxiv.org/abs/2305.10601

    2023 arXiv

  68. [76]

    Di Zhang, Xiaoshui Huang, Dongzhan Zhou, Yuqiang Li, and Wanli Ouyang. 2024. Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self- refine with LLaMa-3 8B. arXiv:2406.07394 [cs.AI] https://arxiv.org/abs/2406.07394

    2024 arXiv

  69. [77]

    Yifan Zhang, Yang Yuan, and Andrew Chi-Chih Yao. 2024. Meta Prompting for AI Systems. arXiv:2311.11482 [cs.AI] https://arxiv.org/abs/2311.11482

    2024 arXiv

  70. [78]

    Chi, Quoc V Le, and Denny Zhou

    Huaixiu Steven Zheng, Swaroop Mishra, Xinyun Chen, Heng-Tze Cheng, Ed H. Chi, Quoc V Le, and Denny Zhou. 2024. Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models. In The Twelfth International Conference on Learning Representations. https://openreview....

    2024

  71. [79]

    Aojun Zhou, Ke Wang, Zimu Lu, Weikang Shi, Sichun Luo, Zipeng Qin, Shaoqing Lu, Anya Jia, Linqi Song, Mingjie Zhan, and Hongsheng Li. 2024. Solving Chal- lenging Math Word Problems Using GPT-4 Code Interpreter with Code-based Self-Verification. In The Twelfth International Con...

    2024

  72. [80]

    Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc V Le, and Ed H. Chi. 2023. Least-to-Most Prompting Enables Complex Reasoning in Large Language Models. In The Eleventh International Conference on ...

    2023

  73. [81]

    Yanqi Zhou, Tao Lei, Hanxiao Liu, Nan Du, Yanping Huang, Vincent Zhao, An- drew M Dai, Quoc V Le, James Laudon, et al. 2022. Mixture-of-experts with expert choice routing. Advances in Neural Information Processing Systems 35 (2022), 7103–7114

    2022

  74. [2001]

    The Knowledge Engineering Review 16, 3 (2001), 277–284

    Learning in multi-agent systems. The Knowledge Engineering Review 16, 3 (2001), 277–284. doi:10.1017/S0269888901000170

    2001 doi

  75. [2003]

    In 11th IEEE/ACM In- ternational Symposium on Modeling, Analysis and Simulation of Computer Telecom- munications Systems, 2003

    Large-scale network simulation: how big? how fast?. In 11th IEEE/ACM In- ternational Symposium on Modeling, Analysis and Simulation of Computer Telecom- munications Systems, 2003. MASCOTS 2003. 116–123. doi:10.1109/MASCOT.2003. 1240649

    2003 doi

  76. [2013]

    InProceedings of the 1st ACM SIGSIM Conference on Principles of Advanced Discrete Simulation (Montr©al, Québec, Canada) (SIGSIM PADS ’13)

    Warp speed: executing time warp on 1,966,080 cores. InProceedings of the 1st ACM SIGSIM Conference on Principles of Advanced Discrete Simulation (Montr©al, Québec, Canada) (SIGSIM PADS ’13). Association for Computing Machinery, New York, NY, USA, 327–336. doi:10.1145/2486092.2486134

  77. [2014]

    In Proceedings of SSST-8, Eighth Workshop on Syntax, Semantics and Structure in Statistical Translation, Dekai Wu, Marine Carpuat, Xavier Carreras, and Eva Maria Vecchi (Eds.)

    On the Properties of Neural Machine Translation: Encoder–Decoder Ap- proaches. In Proceedings of SSST-8, Eighth Workshop on Syntax, Semantics and Structure in Statistical Translation, Dekai Wu, Marine Carpuat, Xavier Carreras, and Eva Maria Vecchi (Eds.). Association for Compu...

    doi

  78. [2017]

    In Proceedings of the 2017 Conference on Empirical Meth- ods in Natural Language Processing

    Supervised Learning of Universal Sentence Representations from Natural Language Inference Data. In Proceedings of the 2017 Conference on Empirical Meth- ods in Natural Language Processing . Association for Computational Linguistics, Copenhagen, Denmark, 670–680. doi:10.18653/v...

    2017 doi

  79. [2023]

    arXiv:2311.06330 [cs.AI] https://arxiv.org/abs/2311.06330

    Smart Agent-Based Modeling: On the Use of Large Language Models in Computer Simulations. arXiv:2311.06330 [cs.AI] https://arxiv.org/abs/2311.06330

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.