Pith. sign in

REVIEW 3 major objections 5 minor 28 references

TrendSim: Simulating Trending Topics in Social Media Under Poisoning Attacks with LLM-based Multi-agent System

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read TrendSim is an LLM-based multi-agent system that simulates trending-topic discussions under poisoning attacks and derives quantitative conclusions about user psychology.

desk verdict TrendSim is a novel trending-topic simulation framework, but its psychological findings are written into the reflection prompt rather than measured. read the letter →

arxiv 2412.12196 v1 pith:CWPGJVPE submitted 2024-12-14 cs.SI cs.AI

Zeyu Zhang , Jianxun Lian , Chen Ma , Yaning Qu , Ye Luo , Lei Wang , Rui Li , Xu Chen
show 4 more authors
Yankai Lin Le Wu Xing Xie Ji-Rong Wen
This is my paper · ORCID
classification cs.SIcs.AI
keywords LLMmulti-agentsimulationtrendingtopicspoisoningattackssocialmediauserpsychologymisinformationcontentcensorship
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

TrendSim is an LLM-based multi-agent simulation designed to reproduce the lifecycle of a trending topic on a social platform—users arrive over time, browse a hub page, comment, like, and reply—and then to inject prototype-based attackers who post antisocial, trolling, or rumor comments. The paper's central claim is that this simulation is realistic and efficient enough to study poisoning attacks that would be impractical to test on real platforms. Using 1,000 user agents distilled from real public posts and 10 trending topics, the authors report that attacks lower users' emotion and social-confidence scores, that positive-sentiment topics suffer the largest drops, that users interested in society are most susceptible, and that LLM-based content censorship measurably mitigates the damage. If the claim holds, TrendSim gives researchers a sandbox for estimating the harm of poisoning attacks and testing defenses before deployment.

What carries the argument

The load-bearing mechanism is the three-level memory of each user agent—long-term memory holding a distilled profile, short-term memory tracking emotion, opinion, and social confidence scores, and flash memory storing the current impression—combined with a reflection process that updates the short-term scores after every interaction. A perception module converts observations into impressions via an LLM, an action module chooses likes, comments, reposts, or replies, and the ranking of comments and the visibility of the centralized topic page make user actions feed back into later observations. Prototype-based attackers generate malicious comments conditioned on a target type (antisocial, trolling, rumor) and the current observation, and the whole system runs on a temporal priority queue so that all interactions are timestamp-ordered.

What would settle it

Run the same TrendSim pipeline but with the reflection prompt's directional constraint reversed or removed (negative content raises or does not change the scores). If the attack-impact curves in Problems 1–4 remain largely unchanged, the reported psychological effects are artifacts of the prompt. A complementary field check would compare simulated emotion and social-confidence trajectories against real user self-reports or platform log data from actual trending-topic cycles; if real users' scores do not move with comment sentiment in the direction the prompt assumes, the results do not transfer.

Watch

Extended reading notes

Core claim

On its own terms, TrendSim establishes that an LLM-driven multi-agent system can simulate trending-topic discussions under poisoning attacks in a way that produces interpretable, quantitative conclusions about user psychology. The discovery is two-fold: the architecture—time-aware access modeled by a rise-peak-decay probability curve, centralized hub dissemination instead of peer-to-peer networks, and agents with perception, memory, and action modules—yields behavior scores comparable to or better than vanilla-LLM and human baselines; and the simulation experiments reveal that poisoning attacks decrease average emotion and social confidence, with the largest absolute effects on positive-sentiment topics and on users interested in social issues, that attacker impact is not proportional to attacker fraction, and that content censorship reduces but does not eliminate the damage.

Load-bearing premise

The load-bearing premise is that an LLM can update a simulated user's emotion and social confidence in a psychologically faithful way, specifically that negative content lowers those scores and positive content raises them; that directional rule is written into the reflection prompt, and every psychological finding in the paper inherits it.

Editorial extensions

If this is right

  • Researchers can estimate the psychological damage of a poisoning attack on a specific trending topic before deploying countermeasures, using the same simulation pipeline.
  • Platforms should prioritize content censorship on positive-sentiment trending topics, since the simulations show those suffer the largest drops in emotion and social confidence.
  • Because attack impact is not proportional to attacker fraction—10% attackers can cause outsized effects—defense systems should not wait for high attack prevalence before intervening.
  • Users whose profiles indicate interest in society are the most susceptible group in the simulations, suggesting targeted protection for that segment.
  • LLM-based content censorship lowers but does not eliminate emotion and confidence losses, so it should be treated as mitigation rather than a complete defense.

Reading between the lines

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

  • If the directional psychology update rule in the reflection prompt is the true driver of the attack-impact results, then a natural next step is field validation: comparing the simulated emotion and social-confidence trajectories against real user self-reports or platform log data from actual trending-topic cycles.
  • The time-aware access function could be reused to study other hub-based content phenomena, such as live event coverage or product-launch threads, where a single page concentrates attention for a few hours.
  • A testable extension would be to reverse or remove the prompt's instruction that negative content lowers scores and positive content raises them; unchanged attack-impact curves would reveal that the psychology results are artifacts of the prompt rather than emergent agent behavior.
  • The non-monotonic effect of attacker fraction hints that there may be a saturation or desensitization dynamic worth probing with formal models, but the paper itself does not establish that mechanism.
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

3 major / 5 minor

Summary. The paper proposes TrendSim, an LLM-based multi-agent system for simulating trending topics on social media under poisoning attacks. The system includes a time-aware interaction mechanism with a piecewise probability distribution for user arrival, centralized message dissemination through a hub, an interactive page model, LLM-based user agents with perception/memory/action modules, and prototype-based attacker agents (antisocial, trolling, rumor). The authors evaluate user agents, attacker agents, the multi-agent system, and simulation efficiency, then use the simulator to study four problems: the negative psychological impact of poisoning attacks, dynamic changes over time, susceptibility by user group, and effectiveness of content censorship. The central claims are that TrendSim is effective and that the simulation experiments reveal how poisoning attacks affect user psychology.

Significance. If the central claims held, TrendSim would be a useful testbed for studying poisoning attacks on trending topics and for evaluating defenses. The paper makes a genuine attempt to address an underexplored scenario, and the design choices—centralized dissemination, time-aware scheduling, and prototype-based attackers—are reasonable abstractions. The presentation is generally clear, and the appendices provide detailed prompts and the G0/G1 smoothness proofs for the time-distribution function, which support replication of the mechanism. However, the significance is heavily conditional: the main empirical conclusions about psychological impact are compromised by a circularity in the reflection prompt, and the effectiveness evidence is inconclusive given the reported variability. The paper does not release code or data, which limits the reproducibility of the numerical results.

major comments (3)
  1. [Appendix D.3, Section 3.3.2, Section 5 (Problems 1-4)] The reflection prompt in Appendix D.3 instructs the LLM: 'This should reflect the change, as the character's psychological conditions are influenced by the information they browse, with positiveness increasing and negativity decreasing.' Since the attacker agents (Section 3.4) are defined to generate negative, antisocial, or rumor-based comments, the observed drops in Emotion and Social Confidence under PA-10/30/50 in Table 5 and Figures 3-4 are a direct consequence of this instruction rather than an emergent or empirically validated psychological response. The 'Psychology Consistency' evaluation in Appendix G.1 rewards post-browsing changes in the expected direction, so high scores on that metric confirm obedience to the instruction, not external validity. Consequently, the conclusions drawn in Problems 1-4 about negative impacts and user susceptibility are not supported as findings about real social-media users; they are artifacts of the prompt design. The Limitations section acknowledges assumptions generally, but it does not disclose that the primary outcome variables are instruction-injected.
  2. [Table 1 and Table 6] Table 1 does not establish the claimed effectiveness of TrendSim for user simulation. On average Behavior Consistency, TrendSim scores 0.915, essentially tied with the vanilla GLM-4 baseline at 0.916, and the error bars in Table 6 (e.g., TrendSim 0.121 under the GPT-4 evaluator) are large and overlapping with baselines. The average Psychology Consistency score of 0.826 is only 0.002 above GLM-4 (0.824) and within the uncertainty of the evaluation. Thus the statement in Section 6 that 'Our evaluations show the effectiveness and efficiency of TrendSim' is not supported by the reported evidence for the user-agent component.
  3. [Section 5, Problem 1, Table 5] The claim that 'the positive group is affected the most' and that 'fewer attackers can probably cause larger impacts' is not supported by the data. For the Positive group, Emotion under SE is 0.886±0.057, while PA-10 is 0.812±0.145, PA-30 is 0.819±0.081, and PA-50 is 0.813±0.048; the standard deviations overlap broadly, and no significance tests are reported. With only 10 trending topics (Section 4.1), the trend of decreases across attack proportions cannot be distinguished from noise. The same issue affects Problem 2's claims of a 'sharp decrease in the middle of time' from curves without confidence intervals, and Problem 4's censorship conclusions in Appendix C.2, where PA-50-CS values are within one standard deviation of PA-50 for most groups.
minor comments (5)
  1. [Section 3.1] Typo: 'express their altitudes' should be 'express their attitudes'.
  2. [Table 5] The group label 'Netural' should be 'Neutral'.
  3. [Appendix F.1 and F.2] The text lists 'GLM-4, GLM-4, and Llama-3' with a duplicate entry; one of the GLM-4 mentions should likely be a different model (e.g., GLM-3-turbo or GPT-4).
  4. [Appendix G.2] Typo: 'topoc' appears twice and should be 'topic'.
  5. [Appendix A] The hyper-parameters A, alpha, and T_m are introduced without a fitting procedure or sensitivity analysis; a sentence describing how they are set for the 10 topics would improve reproducibility.

Circularity Check

2 steps flagged · score 8.0 of 10

Problem 1-4 psychological findings are encoded in the Appendix D.3 reflection prompt, which instructs agents that browsed content raises or lowers emotion and social confidence in exactly the direction attackers' negative comments push.

  1. self definitional [Appendix D.3 (Memory Module prompt); outcome variables Emotion and Social Confidence used in Section 5 Problems 1-4]
    "Please base on your previous psychological conditions, combined with the current impression and actions, output a percentage that objectively represents the positiveness of your current emotion. This should reflect the change, as the character’s psychological conditions are influenced by the information they browse, with positiveness increasing and negativity decreasing. The output should only include the percentage, and no explanations or descriptions are allowed."

    Emotion and Social Confidence are the outcome variables of Problems 1-4. The reflection prompt explicitly instructs the LLM to lower them when browsed information is negative and raise them when positive. Antisocial/trolling/rumor attackers (Section 3.4) are defined to produce negative, antisocial, or rumor-based comments, so raising the attacker proportion injects more negative information; the prompt then forces the recorded emotion and confidence to drop. Tables 5 and 9 and Figures 3-4 therefore report the prompt's built-in direction as an empirical discovery about poisoning attacks, rather than an emergent psychological response.

  2. self definitional [Appendix G.1 (Psychology Consistency evaluation prompt); Table 1]
    "After browsing through the trending topic, his/her psychological condition changes as follows: His/her emotional positiveness score is [Emotion]/1.0, and the social confidence score is[Social Confidence]/1.0. Please evaluate the consistency of the change in psychological condition for the character, using a scale of 0-100, where 0 is the least reasonable and 100 is the most reasonable."

    This is the paper's validation that user agents reflect psychology consistently. The evaluator receives the pre- and post-browsing scores and is asked to judge whether the change matches the character and content, with no external psychological ground truth. Because the post-browsing scores were generated by the Appendix D.3 prompt that already instructs 'positiveness increasing and negativity decreasing,' a high Psychology Consistency score (TrendSim 0.826 in Table 1) largely certifies obedience to the injected causal direction. The evaluation metric and the measured outcome thus share the same assumption, making the validation circular rather than independent.

full rationale

The framework's engineering contributions--time-aware interaction, centralized message dissemination, and the interactive page system--are described independently of the psychological outcome claims, and the attacker-comment quality and efficiency evaluations compare against external baselines. However, the paper's central empirical conclusions in Section 5 (Problems 1-4) about how poisoning attacks affect users' Emotion and Social Confidence are not emergent. The reflection prompt in Appendix D.3 instructs the agent that psychological conditions are influenced by browsed information 'with positiveness increasing and negativity decreasing,' and the same instruction is applied to Social Confidence. Since the attacker prototypes in Section 3.4 are designed to generate negative, antisocial, trolling, or rumor-based comments, increasing the attacker proportion makes the observed content more negative, and the prompt then compels the recorded emotion and confidence scores to fall. Tables 5 and 9 and Figures 3-4 thus present the prompt's own causal assumption as a simulated discovery. The Psychology Consistency evaluation in Appendix G.1 is similarly self-referential: it scores whether the prompted change is consistent with the content rather than comparing against real-user psychological data. These are not independent findings; they are the model's input assumptions returned as output. Non-circular components (e.g., comment consistency/concealment, rationality/diversity, time cost) do not rescue the main psychological-impact claims. Score 8 because the central Problem 1-4 results are forced by definition/prompt; the paper is not entirely devoid of independent engineering content, so 10 would overstate.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central claim depends on several hand-set hyperparameters (arrival curve, initial psychology), an unvalidated psychological reflection rule, and LLM-as-judge scoring. The most consequential is the reflection rule, because it forces the direction of the observed 'impact' and therefore determines all four experimental conclusions.

free parameters (4)
  • A (breaking-degree parameter) = not reported
    Controls steepness of user arrival curve in P(t); chosen per topic without reported values.
  • alpha (curve-shape hyperparameter) = not reported
    Adjusts width of traffic peak; no values given for any simulation run.
  • T_m (peak time hyperparameter) = not reported
    Peak arrival time; set per trending topic, value never stated.
  • Initial emotion and social confidence scores = not reported
    Starting psychological values for agents are not documented, yet all 'impact' statistics are derived from changes around these starting points.
assumptions (5)
  • domain assumption Users arrive according to the piecewise P(t) distribution (exponential, parabola, power-law).
    Section 3.2.1 cites Lerman and Hogg but does not validate the fit for the 10 topics used.
  • ad hoc to paper Negative content decreases emotion/social confidence and positive content increases them; the reflection prompt instructs the model to do exactly this.
    Appendix D.3: 'This should reflect the change, as the character's psychological conditions are influenced by the information they browse, with positiveness increasing and negativity decreasing.'
  • domain assumption LLM-generated scores are treated as valid measures of consistency, rationality, diversity, and concealment.
    Appendix G; scores differ across GPT-4, GLM-4, and Llama-3 evaluators, and only one human volunteer is used.
  • domain assumption Text-only simulation captures the essence of trending-topic dynamics.
    Stated in Limitations; multimodal content such as fake images is excluded.
  • ad hoc to paper Each session has a maximum number of interactions.
    Section 3.2.3; the cap is arbitrary and not calibrated to real usage.
invented entities (1)
  • Prototype-based attacker agents (antisocial, trolling, rumor)
    purpose: Generate topic-specific malicious comments to simulate poisoning attacks
    The three attacker prototypes are author-defined prompt templates with no validation against real-world poisoning behavior; concealment/consistency are scored only by LLMs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TrendSim: Simulating Trending Topics in Social Media Under Poisoning Attacks with LLM-based Multi-agent System." pith.science (2026). https://pith.science/paper/CWPGJVPE

@misc{pith2026241212196,
  author       = {Pith},
  title        = {Pith review of: TrendSim: Simulating Trending Topics in Social Media Under Poisoning Attacks with LLM-based Multi-agent System},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CWPGJVPE}},
  note         = {Machine review of arXiv:2412.12196}
}
read the original abstract

Trending topics have become a significant part of modern social media, attracting users to participate in discussions of breaking events. However, they also bring in a new channel for poisoning attacks, resulting in negative impacts on society. Therefore, it is urgent to study this critical problem and develop effective strategies for defense. In this paper, we propose TrendSim, an LLM-based multi-agent system to simulate trending topics in social media under poisoning attacks. Specifically, we create a simulation environment for trending topics that incorporates a time-aware interaction mechanism, centralized message dissemination, and an interactive system. Moreover, we develop LLM-based human-like agents to simulate users in social media, and propose prototype-based attackers to replicate poisoning attacks. Besides, we evaluate TrendSim from multiple aspects to validate its effectiveness. Based on TrendSim, we conduct simulation experiments to study four critical problems about poisoning attacks on trending topics for social benefit.

Figures

Figures reproduced from arXiv: 2412.12196 by the authors.

Figure 1
Figure 1. The framework of TrendSim for simulating trending topics in social media. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The framework of LLM-based user agents in TrendSim. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The user’s psychological conditions in social [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The user psychological conditions in different groups of simulation experiments. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Compare f(t) with real-world populations. 11 [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: The user’s psychological conditions over time [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: The user’s psychological conditions over time [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: The user’s psychological conditions over time [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 10 canonical work pages

  1. [1]

    Esma A \" meur, Nicol \'a s D \' az Ferreyra, and Hicham Hage. 2019. Manipulation and malicious personalization: exploring the self-disclosure biases exploited by deceptive attackers on social media. Frontiers in artificial intelligence, 2:26

    work page 2019

  2. [2]

    Brian A Barsky and Tony D DeRose. 1989. Geometric continuity of parametric curves: three equivalent characterizations. IEEE Computer Graphics and Applications, 9(6):60--69

    work page 1989

  3. [3]

    Erica J Briscoe, D Scott Appling, and Heather Hayes. 2014. Cues to deception in social media communications. In 2014 47th Hawaii international conference on system sciences, pages 1435--1443. IEEE

    work page 2014

  4. [4]

    Sneha Chinivar, MS Roopa, JS Arunalatha, and KR Venugopal. 2022. Online offensive behaviour in socialmedia: Detection approaches, comprehensive review and future directions. Entertainment Computing, page 100544

    work page 2022

  5. [5]

    Zhengxiao Du, Yujie Qian, Xiao Liu, Ming Ding, Jiezhong Qiu, Zhilin Yang, and Jie Tang. 2022. Glm: General language model pretraining with autoregressive blank infilling. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 320--335

    2022

  6. [6]

    Chen Gao, Xiaochong Lan, Nian Li, Yuan Yuan, Jingtao Ding, Zhilun Zhou, Fengli Xu, and Yong Li. 2023 a . Large language models empowered agent-based modeling and simulation: A survey and perspectives. arXiv preprint arXiv:2312.11970

    arXiv 2023

  7. [7]

    Chen Gao, Xiaochong Lan, Zhihong Lu, Jinzhu Mao, Jinghua Piao, Huandong Wang, Depeng Jin, and Yong Li. 2023 b . S ^3 : Social-network simulation system with large language model-empowered agents. arXiv preprint arXiv:2307.14984

    arXiv 2023

  8. [8]

    Wenyue Hua, Lizhou Fan, Lingyao Li, Kai Mei, Jianchao Ji, Yingqiang Ge, Libby Hemphill, and Yongfeng Zhang. 2023. War and peace (waragent): Large language model-based multi-agent simulation of world wars. arXiv preprint arXiv:2311.17227

    arXiv 2023

Show all 28 references
  1. [9]

    Nitika Khurana, Sudip Mittal, Aritran Piplai, and Anupam Joshi. 2019. Preventing poisoning attacks on ai based threat intelligence systems. In 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing (MLSP), pages 1--6. IEEE

    2019

  2. [10]

    Grgur Kova c , R \'e my Portelas, Peter Ford Dominey, and Pierre-Yves Oudeyer. 2023. The socialai school: Insights from developmental psychology towards artificial socio-cultural agents. arXiv preprint arXiv:2307.07871

    2023 arXiv

  3. [11]

    Rakesh Singh Kunwar and Priyanka Sharma. 2016. Social media: A new vector for cyber attack. In 2016 International Conference on Advances in Computing, Communication, & Automation (ICACCA)(Spring), pages 1--5. IEEE

    2016

  4. [12]

    Kristina Lerman and Tad Hogg. 2010. Using a model of social dynamics to predict popularity of news. In Proceedings of the 19th international conference on World wide web, pages 621--630

    2010

  5. [13]

    Jiaju Lin, Haoran Zhao, Aochi Zhang, Yiting Wu, Huqiuyue Ping, and Qin Chen. 2023. Agentsims: An open-source sandbox for large language model evaluation. arXiv preprint arXiv:2308.04026

    2023 arXiv

  6. [14]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 3...

    2022

  7. [15]

    O'Brien, Carrie J

    Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In In the 36th Annual ACM Symposium on User Interface Software and Technology (UIST '23, UIST '23,...

    2023

  8. [16]

    Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, et al. 2023. ToolLLM : Facilitating large language models to master 16000+ real-world apis. arXiv preprint arXiv:2307.16789

    2023 arXiv

  9. [17]

    Noah Shinn, Federico Cassano, Beck Labash, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. arXiv preprint arXiv:2303.11366

    2023 arXiv

  10. [18]

    Robert L Solso and Jerome Kagan. 1979. Cognitive psychology. Houghton Mifflin Harcourt P

    1979

  11. [19]

    Peter van Emde Boas, Robert Kaas, and Erik Zijlstra. 1976. Design and implementation of an efficient priority queue. Mathematical systems theory, 10(1):99--127

    1976

  12. [20]

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. 2023 a . A survey on large language model based autonomous agents. arXiv preprint arXiv:2308.11432

    2023 arXiv

  13. [21]

    Lei Wang, Jingsen Zhang, Hao Yang, Zhiyuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Ruihua Song, Wayne Xin Zhao, Jun Xu, Zhicheng Dou, Jun Wang, and Ji-Rong Wen. 2023 b . https://arxiv.org/abs/2306.02552 When large language model based agent meets user behavior anal...

    2023 arXiv

  14. [22]

    Zihao Wang, Shaofei Cai, Anji Liu, Xiaojian Ma, and Yitao Liang. 2023 c . Describe, explain, plan and select: Interactive planning with large language models enables open-world multi-task agents. arXiv preprint arXiv:2302.01560

    2023 arXiv

  15. [23]

    Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, Yiwen Ding, Boyang Hong, Ming Zhang, Junzhe Wang, Senjie Jin, Enyu Zhou, et al. 2023. The rise and potential of large language model based agents: A survey. arXiv preprint arXiv:2309.07864

    2023 arXiv

  16. [24]

    Zeyu Zhang, Xiaohe Bo, Chen Ma, Rui Li, Xu Chen, Quanyu Dai, Jieming Zhu, Zhenhua Dong, and Ji-Rong Wen. 2024. A survey on the memory mechanism of large language model based agents. arXiv preprint arXiv:2404.13501

    2024 arXiv

  17. [25]

    Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al. 2023. A survey of large language models. arXiv preprint arXiv:2303.18223

    2023 arXiv

  18. [26]

    Xizhou Zhu, Yuntao Chen, Hao Tian, Chenxin Tao, Weijie Su, Chenyu Yang, Gao Huang, Bin Li, Lewei Lu, Xiaogang Wang, et al. 2023. Ghost in the minecraft: Generally capable agents for open-world enviroments via large language models with text-based knowledge and memory. arXiv pr...

    2023 arXiv

  19. [27]

    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...

  20. [28]

    write newline

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

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.