Pith. sign in

REVIEW 3 major objections 5 minor 50 references

Too Human to Model:The Uncanny Valley of LLMs in Social Simulation -- When Generative Language Agents Misalign with Modelling Principles

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper argues that LLM agents are too expressive, detailed, and intractable to serve the abstraction and interpretability that social simulation requires.

desk verdict A clearly argued position paper that gives a useful five-dilemma framing for the mismatch between LLM agents and explanatory social simulation; the 'intrinsic' claim outstrips the evidence, but the taxonomy and comfort zones are citable contributions. read the letter →

arxiv 2507.06310 v1 pith:Z5Z3VBIF submitted 2025-07-08 cs.CY cs.AIcs.MA

Yongchao Zeng , Calum Brown , Mark Rounsevell This is my paper
classification cs.CYcs.AIcs.MA
keywords largelanguagemodelssocialsimulationagent-basedmodellinguncannyvalleyabstractioninterpretabilityBassdiffusionmodelemergence
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that using large language models as the cognitive engine of social simulation agents conflicts with modelling's core epistemic purpose: abstraction. Because LLM agents produce fluent, contextually rich, turn-by-turn dialogue, they inject a level of detail that blurs the distinction between micro-level behaviour and macro-level emergence rather than clarifying it. The authors demonstrate this through a thought-experiment redesign of the Bass diffusion model, where each necessary modelling decision—time scale, conversation control, prompt rules, role evolution, aggregation—surfaces a dilemma. Together the five dilemmas push LLM agents into an 'uncanny valley': too artificial to pass as human behaviour and too verbose to function as an abstract mechanism. The paper concludes that LLM agents belong in simulation niches where linguistic nuance matters and system-level emergence is not the object of study.

What carries the argument

The load-bearing device is a model-building thought experiment: conversion of the classic Bass diffusion model into an LLM-driven agent-based simulation. The Bass model's compact equation, $f(t) = [p + qF(t)][1 - F(t)]$, summarises diffusion as the product of external influence ($p$) and imitation ($q$); the thought experiment replaces the transparent IF-THEN adoption rule of the agent-based version with prompted LLM agents who converse, decide, and accumulate role histories. Each conversion step forces a design choice, and each choice generates one of the five dilemmas, so the thought experiment itself functions as a diagnostic instrument that localises where realism and abstraction collide.

What would settle it

Implement two parallel agent-based diffusion models—one rule-based with explicit $p$ and $q$, one LLM-driven with perfect memory and robust role alignment—and check whether the LLM version's aggregated adoption curve can be decomposed into identifiable, stable causal parameters through sensitivity analysis. If the LLM-based model permits reliable parameter identification and causal tracing from dialogue to curve, the claimed epistemic uncanny valley would be refuted.

Watch

Extended reading notes

Core claim

LLM agents are, in the authors' phrasing, 'too human to model': their realism is not a neutral enhancement but an epistemic liability in simulations built to explain social dynamics. Working through the construction of an LLM-based variant of the Bass diffusion model, the paper identifies five structural dilemmas—a temporal resolution mismatch between yearly abstract steps and conversational time; the need to intervene in conversations without destroying spontaneous output; the temptation to inject rule-like instructions into prompts while keeping dialogue natural; the tension between robust role alignment and credible role evolution; and the opacity of emergence in verbose textual logs. The net effect is a paradox: the more human the agents become, the less the model can reveal about the social mechanisms it represents. The paper therefore restricts the legitimate use of LLM agents to settings where system-level emergence is not the focus, linguistic nuances and meaning are central, interactions occur in natural time, and stable role identity outweighs long-term behavioural evolution.

Load-bearing premise

The load-bearing premise is that the five dilemmas are intrinsic to LLM agents in social simulation and would persist even if every current technical limitation—memory, output stability, role alignment—were solved, so that better scaffolding could not make the uncanny valley disappear.

Editorial extensions

If this is right

  • LLM agents should not replace rule-based agents in simulations whose purpose is explaining emergent system-level patterns, because narrative-rich output obscures the mechanisms that such models exist to reveal.
  • The realism of an agent is not an unqualified virtue; in mechanism-oriented simulation, added expressiveness can reduce interpretability, calibration, and the ability to isolate causal variables.
  • The five dilemmas provide a practical checklist: before embedding LLM agents, modellers should ask whether their research question requires abstract time steps, intervention-free spontaneity, rule-free decisions, long-term role evolution, or transparent emergence.
  • LLM agents are best confined to 'comfort zones' such as social learning, situated role play, deliberative discourse, and human-in-the-loop prototyping, where linguistic fidelity is the goal rather than system-level prediction or explanation.

Reading between the lines

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

  • If the five dilemmas are intrinsic rather than engineering artefacts, then even a hypothetically perfect memory-and-alignment scaffold would not rehabilitate LLM agents for explanation-oriented simulation; the right test is a maximal-scaffolding implementation that still fails parameter identifiability.
  • One way to sharpen the uncanny valley into a measurable quantity is to vary the linguistic detail of agent outputs and track how quickly causal variables such as $p$ and $q$ become non-identifiable from aggregate dynamics; the paper suggests such a trade-off without computing it.
  • Read this way, LLM-based and rule-based simulations are different epistemic genres—narrative exploration versus mechanism explanation—and will need separate validation criteria rather than a single realism scale.
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 argues that LLM-based agents, despite their ability to generate fluent and contextually rich behaviour, create a fundamental epistemic mismatch when used as components of social simulations whose purpose is to explain emergent system-level patterns. The authors support this claim through a modelling thought experiment in which the classic Bass diffusion model is rebuilt using LLM agents. The exercise yields five dilemmas: a temporal resolution mismatch between natural conversation and abstract time steps; the need for intervention in conversations; the temptation to reintroduce rule-like instructions; a tension between role alignment and role evolution; and the obscuring of emergence by verbose textual outputs. Together, the authors argue, these dilemmas push LLM agents into an 'uncanny valley' that is neither abstract enough to clarify mechanisms nor realistic enough to represent human behaviour. The paper closes by proposing 'comfort zones' where LLM agents are better suited and calls for repositioning LLM agents in the broader ecosystem of social simulation.

Significance. The paper addresses a timely and important question: whether the growing use of LLM agents in social simulation advances or undermines the explanatory aims of modelling. If the central claim is accepted, the paper provides a valuable caution against treating LLM agents as drop-in replacements for rule-based agents in mechanism-oriented models, and it offers a useful vocabulary (the five dilemmas, the uncanny valley) for discussing these issues. A notable strength is the paper's transparency: it explicitly labels the exercise as a hypothetical thought experiment, states its assumptions as it goes, and avoids a blanket dismissal of LLM agents by identifying positive conditions for their use. The main significance is therefore conditional: the argument is coherent and clearly structured, but the evidence is a single worked example, and the modal and general phrasing of the conclusion ('too human to model') exceeds what the thought experiment alone can establish. The paper is best read as a conceptual critique or position paper, and its value will depend on whether its load-bearing assumptions can be supported or appropriately qualified.

major comments (3)
  1. [Section 3, Dilemma V] The central claim that a 'more profound methodological and epistemological tension' exists 'even if all these technical limitations are solved' is asserted rather than demonstrated. The thought experiment in Section 3 generates each dilemma from a prior design choice: Dilemma I assumes yearly conversations with 'cognitive compression'; the response to Dilemma I creates Dilemma II; the response to Dilemma II creates Dilemma III; and so on. Alternative architectures that appear in the LLM-agent literature, such as hybrid systems with structured state variables tracking adoption status alongside narrative generation, or asynchronous event-based scheduling, are not considered. If such architectures avoid one or more dilemmas, the 'uncanny valley' is an artefact of the chosen modelling style rather than a fundamental property of LLM-based simulation. The authors should either weaken the universal phrasing ('too human to model') to a scope-restricted claim about the examined modelling choices, or provide a non-circular argument for why all alternative framings are excluded.
  2. [Section 3, Dilemma V] The conclusion that system-level emergence is obscured by verbose textual output ('a sea of words, a desert of meaning') rests on the assumption that the analyst uses the text only to count '1' and '0' and to read logs. The paper does not discuss computational text-analysis methods, such as structured event extraction, topic modeling, embedding-based aggregation, or LLM-assisted coding, that could recover quantitative variables from narrative output. If such methods can restore interpretability and parameter levers, Dilemma V becomes an engineering challenge rather than an epistemic one. Since this dilemma is the capstone of the uncanny-valley argument, the authors should either demonstrate why text-derived variables cannot support calibration and sensitivity analysis in principle, or reframe the claim as a practical limitation of current practice.
  3. [Section 4, Table 1] The 'LLM Fit Assessment' column in Table 1 assigns qualitative ratings (e.g., Explanation: 'Weak', Prediction: 'Mixed', Social Learning: 'Strong') without a stated rubric, systematic evidence, or citation support per cell. These ratings feed directly into the paper's positive recommendation to reposition LLM agents into comfort zones, so they are not decorative. The authors should either provide explicit criteria for the ratings, support them with a systematic review of applications, or explicitly present them as untested hypotheses requiring future work.
minor comments (5)
  1. [Manuscript front matter] The abstract appears twice in the submitted text: once at the top and again after the author block. Please remove the duplicate.
  2. [Section 2] The sentence 'an individual adopts an innovation due to the influence of social media and the people connecting with that individual' should say 'mass media' to match Bass's terminology and the subsequent paragraph.
  3. [Figure 1] Figure 1 is referenced and captioned but the figure image is not visible in the manuscript text; please ensure the figure is included.
  4. [References] There are several formatting inconsistencies in the reference list, e.g., 'arXiv reprints' for Liu et al. (2024a) and some entries without page ranges; the citation to Kant (2024) appears unrelated to the point about asking questions wisely and should be replaced or removed.
  5. [Section 3, Dilemma II] The structured cue '1'/'0' is mentioned as one option, but it is later assumed during aggregation; please clarify whether such structured outputs are part of the proposed design or extracted from free-form conversation logs.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the thought experiment traces consequences of explicit modelling choices, and no prediction, fitted parameter, or load-bearing self-citation is involved.

full rationale

This paper is a conceptual and argumentative essay rather than an empirical derivation. It does not fit parameters, make quantitative predictions, or test hypotheses against data. The central thesis—that LLM agents are "too human to model"—is supported by a deliberately hypothetical modelling exercise in Section 3 that converts the Bass diffusion model into an LLM-based variant and identifies five dilemmas. The dilemmas are presented as consequences of explicit design choices made along the way, such as assuming "cognitive compression" to proceed past Dilemma I, and the paper openly labels the exercise "hypothetical" and "a modelling thought experiment". In Section 4, the authors explicitly state that the five dilemmas "do not point to any inherent shortcomings of LLMs" but rather demonstrate a mismatch between a flexible tool and certain modelling purposes, which further weakens any reading of the dilemmas as a forced derivation from hidden premises. No equation in the paper is used to predict an outcome that is then fed back as an input; no fitted parameter is renamed as a prediction; and no uniqueness theorem or equivalent result is imported from the authors' prior work. The self-citations that appear (Zeng et al., 2024a, 2025) are used as illustrative application examples or as support for a well-known technical limitation (output instability), and they are not load-bearing for the core claim. Whether the five dilemmas are genuinely intrinsic to LLM-based simulation, or merely artefacts of the particular thought-experiment design, is a question of argumentative support and empirical testing, not circularity. That concern belongs under correctness risk, not under the circularity score.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The paper has no free parameters and no invented entities. It rests on one methodological axiom (models should be abstract and interpretable) and a representativeness assumption (the Bass model illustrates the epistemic strengths of social simulation).

assumptions (2)
  • domain assumption The primary purpose of social simulation is abstraction, simplification, and interpretability for explanation, not realism per se.
    Section 1 and Section 4 argue that realism conflicts with modelling; this evaluative premise is asserted rather than defended.
  • domain assumption The Bass diffusion model is representative of the epistemic strengths and demands of social simulation.
    Section 2 selects Bass as the template; the general conclusions assume that the dilemmas found in converting Bass would arise in other simulation models as well.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Too Human to Model:The Uncanny Valley of LLMs in Social Simulation -- When Generative Language Agents Misalign with Modelling Principles." pith.science (2026). https://pith.science/paper/Z5Z3VBIF

@misc{pith2026250706310,
  author       = {Pith},
  title        = {Pith review of: Too Human to Model:The Uncanny Valley of LLMs in Social Simulation -- When Generative Language Agents Misalign with Modelling Principles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z5Z3VBIF}},
  note         = {Machine review of arXiv:2507.06310}
}
read the original abstract

Large language models (LLMs) have been increasingly used to build agents in social simulation because of their impressive abilities to generate fluent, contextually coherent dialogues. Such abilities can enhance the realism of models. However, the pursuit of realism is not necessarily compatible with the epistemic foundation of modelling. We argue that LLM agents, in many regards, are too human to model: they are too expressive, detailed and intractable to be consistent with the abstraction, simplification, and interpretability typically demanded by modelling. Through a model-building thought experiment that converts the Bass diffusion model to an LLM-based variant, we uncover five core dilemmas: a temporal resolution mismatch between natural conversation and abstract time steps; the need for intervention in conversations while avoiding undermining spontaneous agent outputs; the temptation to introduce rule-like instructions in prompts while maintaining conversational naturalness; the tension between role consistency and role evolution across time; and the challenge of understanding emergence, where system-level patterns become obscured by verbose micro textual outputs. These dilemmas steer the LLM agents towards an uncanny valley: not abstract enough to clarify underlying social mechanisms, while not natural enough to represent realistic human behaviour. This exposes an important paradox: the realism of LLM agents can obscure, rather than clarify, social dynamics when misapplied. We tease out the conditions in which LLM agents are ideally suited: where system-level emergence is not the focus, linguistic nuances and meaning are central, interactions unfold in natural time, and stable role identity is more important than long-term behavioural evolution. We call for repositioning LLM agents in the ecosystem of social simulation for future applications.

Figures

Figures reproduced from arXiv: 2507.06310 by the authors.

Figure 1
Figure 1. The parameters of the baseline curve are p = 0.01 and q = 0.15. Manipulating the values of p and q alters the shape of the diffusion curve: A higher p accelerates early adoption, making the inflection point happen earlier. A higher q amplifies adoption later in the process, steepening the curve after the inflection point. 3. Constructing an LLM-Driven Diffusion Model: A Modelling Thought Experiment To vividly reveal… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 22 canonical work pages

  1. [4]

    Education, Training, Situated Role Play Real-time conversation, role- play fidelity, situated dialogue Classroom simulations (Zhang et al., 2024b), empathic conversation (Wang et al., 2025), clinical role-play (Li et al., 2024c; Louie et al.,

    work page 2025

  2. [5]

    Human-in-the-Loop Systems & Speculative Prototyping Co-creative dialogue, human- guided iteration Design fiction (Wu et al., 2025), speculative prototyping (Tost et al., 2024), decision support systems (Li et al.,

    work page 2025

  3. [6]

    Analysis of LLM-based narrative generation using the agent -based simulation. 2023 15th International Congress on Advanced Applied Informatics Winter (IIAI-AAI-Winter), 284-289.https://doi.org/10.1109/IIAI-AAI-Winter61682.2023.00059 Arslan, M., Ghanem, H., Munawar, S., Cruz, C.,

    arXiv 2023

  4. [7]

    Procedia Computer Science 246, 3781-3790.https://doi.org/10.1016/j.procs.2024.09.178 Bass, F.M.,

    A Survey on RAG with LLMs. Procedia Computer Science 246, 3781-3790.https://doi.org/10.1016/j.procs.2024.09.178 Bass, F.M.,

    doi:10.1016/j.procs.2024.09.178 2024

  5. [10]

    ISPRS International Journal of GEO -information 9, 581.https://doi.org/10.3390/ijgi9100581 Carrasco-Farre, C.,

    Supporting policy design for the diffusion of cleaner technologies: A spatial empirical agent -based model. ISPRS International Journal of GEO -information 9, 581.https://doi.org/10.3390/ijgi9100581 Carrasco-Farre, C.,

    work page doi:10.3390/ijgi9100581

  6. [11]

    arXiv preprint arXiv:2404.09329.https://doi.org/10.48550/arXiv.2404.09329 Chen, Z.,

    Large language models are as persuasive as humans, but how? About the cognitive effort and moral -emotional language of LLM arguments. arXiv preprint arXiv:2404.09329.https://doi.org/10.48550/arXiv.2404.09329 Chen, Z.,

    arXiv doi:10.48550/arxiv.2404.09329

  7. [18]

    OSF Preprints.https://doi.org/10.31219/osf.io/ey8db He, Z., Cao, P., Wang, C., Jin, Z., Chen, Y., Xu, J., Li, H., Jiang, X., Liu, K., Zhao, J.,

    Comparing the persuasiveness of role - playing large language models and human experts on polarized US political issues. OSF Preprints.https://doi.org/10.31219/osf.io/ey8db He, Z., Cao, P., Wang, C., Jin, Z., Chen, Y., Xu, J., Li, H., Jiang, X., Liu, K., Zhao, J.,

    work page doi:10.31219/osf.io/ey8db

  8. [19]

    arXiv preprint arXiv:2403.02959

    AgentsCourt: Building Judicial Decision -Making Agents with Court Debate Simulation and Legal Knowledge Augmentation. arXiv preprint arXiv:2403.02959. https://doi.org/10.48550/arXiv.2403.02959 Horiguchi, I., Yoshida, T., Ikegami, T.,

    arXiv doi:10.48550/arxiv.2403.02959

Show all 50 references
  1. [20]

    arXiv preprint arXiv:2409.00993.https://doi.org/10.48550/arXiv.2409.00993 Jiang, P., Sonne, C., Li, W., You, F., You, S.,

    Evolution of Social Norms in LLM Agents using Natural Language. arXiv preprint arXiv:2409.00993.https://doi.org/10.48550/arXiv.2409.00993 Jiang, P., Sonne, C., Li, W., You, F., You, S.,

    doi arXiv

  2. [21]

    Engineering 40, 202 - 210.https://doi.org/10.1016/j.eng.2024.04.002 Jones, C.R., Bergen, B.K.,

    Preventing the immense increase in the life-cycle energy and carbon footprints of llm -powered intelligent chatbots. Engineering 40, 202 - 210.https://doi.org/10.1016/j.eng.2024.04.002 Jones, C.R., Bergen, B.K.,

    2024 doi

  3. [22]

    arXiv preprint arXiv:2503.23674.https://doi.org/10.48550/arXiv.2503.23674 Kant, I.,

    Large language models pass the turing test. arXiv preprint arXiv:2503.23674.https://doi.org/10.48550/arXiv.2503.23674 Kant, I.,

    doi arXiv

  4. [23]

    Central European Journal of Operations Research 20, 183 - 230.https://doi.org/10.1007/s10100-011-0210-y Knuuttila, T.,

    Agent -based simulation of innovation diffusion: a review. Central European Journal of Operations Research 20, 183 - 230.https://doi.org/10.1007/s10100-011-0210-y Knuuttila, T.,

    doi

  5. [26]

    Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 1-18.https://doi.org/10.1145/3706598.3713133 Lin, L., Wang, L., Guo, J., Wong, K

    From Text to Trust: Empowering AI -assisted Decision Making with Adaptive LLM -powered Analysis. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 1-18.https://doi.org/10.1145/3706598.3713133 Lin, L., Wang, L., Guo, J., Wong, K. -F.,

    2025

  6. [27]

    We Need Structured Output

    Investigating bias in llm -based bias detection: Disparities between llms and human perception. arXiv preprint arXiv:2403.14896.https://doi.org/10.48550/arXiv.2403.14896 Liu, M.X., Liu, F., Fiannaca, A.J., Koo, T., Dixon, L., Terry, M., Cai, C.J., 2024a. " We Need Structured O...

    doi arXiv

  7. [28]

    arXiv preprint arXiv:2308.05374.https://doi.org/10.48550/arXiv.2308.05374 Louie, R., Nandi, A., Fang, W., Chang, C., Brunskill, E., Yang, D.,

    Trustworthy llms: a survey and guideline for evaluating large language models' alignment. arXiv preprint arXiv:2308.05374.https://doi.org/10.48550/arXiv.2308.05374 Louie, R., Nandi, A., Fang, W., Chang, C., Brunskill, E., Yang, D.,

    doi arXiv

  8. [29]

    arXiv preprint arXiv:2407.00870.https://doi.org/10.48550/arXiv.2407.00870 Miller, J.H., Page, S.E.,

    Roleplay-doh: Enabling domain- experts to create llm -simulated patients via eliciting and adhering to principles. arXiv preprint arXiv:2407.00870.https://doi.org/10.48550/arXiv.2407.00870 Miller, J.H., Page, S.E.,

    doi arXiv

  9. [30]

    2024 47th MIPRO ICT and Electronics Convention (MIPRO), 2084 - 2088.https://doi.org/10.1109/MIPRO60963.2024.10569238 Rand, W., Rust, R.T.,

    Hallucinations in llms: Understanding and addressing challenges. 2024 47th MIPRO ICT and Electronics Convention (MIPRO), 2084 - 2088.https://doi.org/10.1109/MIPRO60963.2024.10569238 Rand, W., Rust, R.T.,

    2024

  10. [33]

    Environmental science & technology 57, 3464 -3466

    Risks and benefits of large language models for the environment. Environmental science & technology 57, 3464 -3466. https://doi.org/10.1021/acs.est.3c01106 Rogers, E.M., Singhal, A., Quinlan, M.M.,

    doi

  11. [34]

    arXiv preprint arXiv:2411.06837.https://doi.org/10.48550/arXiv.2411.06837 Shi, Y., Wei, Z., Shahbaz, M., Zeng, Y.,

    Persuasion with Large Language Models: a Survey. arXiv preprint arXiv:2411.06837.https://doi.org/10.48550/arXiv.2411.06837 Shi, Y., Wei, Z., Shahbaz, M., Zeng, Y.,

    doi arXiv

  12. [35]

    Energy Economics 101, 105399.https://doi.org/10.1016/j.eneco.2021.105399 Shi, Y., Zeng, Y., Engo, J., Han, B., Li, Y., Muehleisen, R.T.,

    Exploring the dynamics of low-carbon technology diffusion among enterprises: An evolutionary game model on a two -level heterogeneous social network. Energy Economics 101, 105399.https://doi.org/10.1016/j.eneco.2021.105399 Shi, Y., Zeng, Y., Engo, J., Han, B., Li, Y., Muehleis...

    2021

  13. [36]

    Applied Energy 263, 114641.https://doi.org/10.1016/j.apenergy.2020.114641 Tamoyan, H., Schuff, H., Gurevych, I.,

    Leveraging inter-firm influence in the diffusion of energy efficiency technologies: An agent -based model. Applied Energy 263, 114641.https://doi.org/10.1016/j.apenergy.2020.114641 Tamoyan, H., Schuff, H., Gurevych, I.,

    2020

  14. [37]

    arXiv preprint arXiv:2407.03974.https://doi.org/10.48550/arXiv.2407.03974 Tao, Y., Viberg, O., Baker, R.S., Kizilcec, R.F.,

    Llm roleplay: Simulating human -chatbot interaction. arXiv preprint arXiv:2407.03974.https://doi.org/10.48550/arXiv.2407.03974 Tao, Y., Viberg, O., Baker, R.S., Kizilcec, R.F.,

    doi arXiv

  15. [38]

    PNAS nexus 3, pgae346

    Cultural bias and cultural alignment of large language models. PNAS nexus 3, pgae346. https://doi.org/10.1093/pnasnexus/pgae346 Tonmoy, S., Zaman, S., Jain, V., Rani, A., Rawte, V., Chadha, A., Das, A.,

    doi

  16. [39]

    arXiv preprint arXiv:2401.01313.https://doi.org/10.48550/arXiv.2401.01313 Tost, J., Gohsen, M., Schulte, B., Thomet, F., Kuhn, M., Kiesel, J., Stein, B., Hornecker, E.,

    A comprehensive survey of hallucination mitigation techniques in large language models. arXiv preprint arXiv:2401.01313.https://doi.org/10.48550/arXiv.2401.01313 Tost, J., Gohsen, M., Schulte, B., Thomet, F., Kuhn, M., Kiesel, J., Stein, B., Hornecker, E.,

    doi arXiv

  17. [40]

    Futuring machines: An interactive framework for participative futuring through human -ai collaborative speculative fiction writing. Proceedings of the 6th ACM Conference on Conversational User Interfaces, 1 - 7.https://doi.org/10.1145/3640794.3665904 Wang, L., Ma, C., Feng, X....

  18. [41]

    International Conference on Intelligent Multilingual Information Processing, 205 - 218.https://doi.org/10.1007/978-981-96-5123-8_14 Wang, Y., Zhou, Q., Ledo, D., 2024b

    EmpLLM: Enhancing Empathy in LLMs Through Psychologist Simulation. International Conference on Intelligent Multilingual Information Processing, 205 - 218.https://doi.org/10.1007/978-981-96-5123-8_14 Wang, Y., Zhou, Q., Ledo, D., 2024b. StoryVerse: Towards co-authoring dynamic ...

  19. [42]

    arXiv preprint, arXiv:2502.17878.https://doi.org/10.48550/arXiv.2502.17878 Xie, Q., Feng, Q., Zhang, T., Li, Q., Zhang, Y., Feng, R., Gao, S.,

    Towards Enhanced Immersion and Agency for LLM - based Interactive Drama. arXiv preprint, arXiv:2502.17878.https://doi.org/10.48550/arXiv.2502.17878 Xie, Q., Feng, Q., Zhang, T., Li, Q., Zhang, Y., Feng, R., Gao, S.,

    doi arXiv

  20. [43]

    arXiv preprints, arXiv: 2402.18180.https://doi.org/10.48550/arXiv.2402.18180 Yan, Z., Xiang, Y.,

    Human Simulacra: Benchmarking the Personification of Large Language Models. arXiv preprints, arXiv: 2402.18180.https://doi.org/10.48550/arXiv.2402.18180 Yan, Z., Xiang, Y.,

    doi arXiv

  21. [44]

    Proceedings of the ACM on Human-Computer Interaction 9, 1-44.https://doi.org/10.1145/3711068 Ye, H., Jin, J., Xie, Y., Zhang, X., Song, G.,

    Social life simulation for non-cognitive skills learning. Proceedings of the ACM on Human-Computer Interaction 9, 1-44.https://doi.org/10.1145/3711068 Ye, H., Jin, J., Xie, Y., Zhang, X., Song, G.,

    doi

  22. [45]

    arXiv preprint arXiv:2505.08245.https://doi.org/10.48550/arXiv.2505.08245 Yu, T., Zhang, Y.-F., Fu, C., Wu, J., Lu, J., Wang, K., Lu, X., Shen, Y., Zhang, G., Song, D.,

    Large Language Model Psychometrics: A Systematic Review of Evaluation, Validation, and Enhancement. arXiv preprint arXiv:2505.08245.https://doi.org/10.48550/arXiv.2505.08245 Yu, T., Zhang, Y.-F., Fu, C., Wu, J., Lu, J., Wang, K., Lu, X., Shen, Y., Zhang, G., Song, D.,

    doi

  23. [46]

    arXiv preprint arXiv:2503.14504.https://doi.org/10.48550/arXiv.2503.14504 Zeng, Y., Brown, C., Byari, M., Raymond, J., Schmitt, T., Rounsevell, M., 2024a

    Aligning multimodal llm with human preference: A survey. arXiv preprint arXiv:2503.14504.https://doi.org/10.48550/arXiv.2503.14504 Zeng, Y., Brown, C., Byari, M., Raymond, J., Schmitt, T., Rounsevell, M., 2024a. InsNet-CRAFTY v1. 0: Integrating institutional network dynamics p...

    doi arXiv

  24. [47]

    Earth System Dynamics 16, 423-449.https://doi.org/10.5194/esd-16-423-2025 Zeng, Y., Dong, P., Shi, Y., Wang, L., Li, Y.,

    Exploring the opportunities and challenges of using large language models to represent institutional agency in land system modelling. Earth System Dynamics 16, 423-449.https://doi.org/10.5194/esd-16-423-2025 Zeng, Y., Dong, P., Shi, Y., Wang, L., Li, Y.,

    2025 doi

  25. [48]

    Journal of Cleaner Production 256, 120384.https://doi.org/10.1016/j.jclepro.2020.120384 Zeng, Y., Shi, Y., Shahbaz, M., Liu, Q., 2024b

    Analyzing the co -evolution of green technology diffusion and consumers’ pro -environmental attitudes: An agent -based model. Journal of Cleaner Production 256, 120384.https://doi.org/10.1016/j.jclepro.2020.120384 Zeng, Y., Shi, Y., Shahbaz, M., Liu, Q., 2024b. Scenario -based...

    2020

  26. [50]

    UIUC Spring 2025 CS598 LLM Agent Workshop

    A survey on alignment for large language model agents. UIUC Spring 2025 CS598 LLM Agent Workshop

    2025

  27. [1969]

    Management science 15, 215 - 227.https://doi.org/10.1287/mnsc.15.5.215 Bass, F.M.,

    A new product growth for model consumer durables. Management science 15, 215 - 227.https://doi.org/10.1287/mnsc.15.5.215 Bass, F.M.,

    doi

  28. [2004]

    a new product growth for model consumer durables the bass model

    Comments on “a new product growth for model consumer durables the bass model”. Management science 50, 1833-1840.https://doi.org/10.1287/mnsc.1040.0300 Caprioli, C., Bottero, M., De Angelis, E.,

  29. [2005]

    Journal of Physics: condensed matter 17, S1259.https://doi.org/10.1088/0953-8984/17/14/015 Hackenburg, K., Ibrahim, L., Tappin, B.M., Tsakiris, M.,

    Heterogeneity and feedback in an agent -based market model. Journal of Physics: condensed matter 17, S1259.https://doi.org/10.1088/0953-8984/17/14/015 Hackenburg, K., Ibrahim, L., Tappin, B.M., Tsakiris, M.,

    doi

  30. [2007]

    Cambridge University Press Li, J., Consul, S., Zhou, E., Wong, J., Farooqui, N., Ye, Y., Manohar, N., Wei, Z., Wu, T., Echols, B., 2024a

    Modelling for field biologists and other interesting people. Cambridge University Press Li, J., Consul, S., Zhou, E., Wong, J., Farooqui, N., Ye, Y., Manohar, N., Wei, Z., Wu, T., Echols, B., 2024a. Banishing LLM hallucinations requires rethinking generalization. arXiv preprin...

    doi arXiv

  31. [2010]

    Renewable and sustainable energy reviews 14, 1070 - 1078.https://doi.org/10.1016/j.rser.2009.11.007 Rillig, M.C., Ågerstrand, M., Bi, M., Gould, K.A., Sauerland, U.,

    A review of technology diffusion models with special reference to renewable energy technologies. Renewable and sustainable energy reviews 14, 1070 - 1078.https://doi.org/10.1016/j.rser.2009.11.007 Rillig, M.C., Ågerstrand, M., Bi, M., Gould, K.A., Sauerland, U.,

    2009 doi

  32. [2011]

    International Journal of research in Marketing 28, 181-193.https://doi.org/10.1016/j.ijresmar.2011.04.002 Rao, K.U., Kishore, V.V.N.,

    Agent-based modeling in marketing: Guidelines for rigor. International Journal of research in Marketing 28, 181-193.https://doi.org/10.1016/j.ijresmar.2011.04.002 Rao, K.U., Kishore, V.V.N.,

    2011 doi

  33. [2012]

    Epidemiologic Perspectives & Innovations 9, 1 - 9.https://doi.org/10.1186/1742-5573-9-1 Epstein, J.M.,

    Social network analysis and agent -based modeling in social epidemiology. Epidemiologic Perspectives & Innovations 9, 1 - 9.https://doi.org/10.1186/1742-5573-9-1 Epstein, J.M.,

    doi

  34. [2017]

    Simulating social complexity: A handbook, 39 -58

    Different modelling purposes. Simulating social complexity: A handbook, 39 -58. https://doi.org/10.1007/978-3-319-66948-9_4 El-Sayed, A.M., Scarborough, P., Seemann, L., Galea, S.,

    doi

  35. [2018]

    Topics in cognitive science 10, 745-758.https://doi.org/10.1111/tops.12327 Edmonds, B.,

    A social approach to rule dynamics using an agent‐based model. Topics in cognitive science 10, 745-758.https://doi.org/10.1111/tops.12327 Edmonds, B.,

    doi

  36. [2019]

    Papers in Applied Geography 5, 77-97.https://doi.org/10.1080/23754931.2019.1619193 Claude,

    An agent -based model for information diffusion over online social networks. Papers in Applied Geography 5, 77-97.https://doi.org/10.1080/23754931.2019.1619193 Claude,

    2019

  37. [2020]

    Spatio -temporal modelling of diffusion of electric vehicles and solar photovoltaic panels: an integrated agent -based and artificial neural networks model.https://dspace.lib.cranfield.ac.uk/handle/1826/20849 Comfort Zone Research Focus Application Cases Deliberative Discourse...

    2024

  38. [2021]

    European journal for philosophy of science 11, 65.https://doi.org/10.1007/s13194-021-00374-5 Kokko, H.,

    Epistemic artifacts and the modal dimension of modeling. European journal for philosophy of science 11, 65.https://doi.org/10.1007/s13194-021-00374-5 Kokko, H.,

    doi

  39. [2022]

    Journal of Cleaner Production 367, 132793.https://doi.org/10.1016/j.jclepro.2022.132793 Zhang, Z., Bo, X., Ma, C., Li, R., Chen, X., Dai, Q., Zhu, J., Dong, Z., Wen, J.-R., 2024a

    An agent -based diffusion model for Residential Photovoltaic deployment in Singapore: Perspective of consumers’ behaviour. Journal of Cleaner Production 367, 132793.https://doi.org/10.1016/j.jclepro.2022.132793 Zhang, Z., Bo, X., Ma, C., Li, R., Chen, X., Dai, Q., Zhu, J., Don...

    2022

  40. [2023]

    arXiv preprint arXiv:2303.08774.https://doi.org/10.48550/arXiv.2303.08774 Alderete Peralta, A.,

    Gpt -4 technical report. arXiv preprint arXiv:2303.08774.https://doi.org/10.48550/arXiv.2303.08774 Alderete Peralta, A.,

    doi arXiv

  41. [2024]

    Cooperation, competition, and maliciousness: LLM -stakeholders interactive negotiation. arXiv preprint, arXiv:2309.17234 https://doi.org/10.48550/arXiv.2309.17234 Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F.L., Almeida, D., Altenschmidt, J., Altman, S....

    doi arXiv

  42. [2025]

    EmoCharacter: Evaluating the Emotional Fidelity of Role -Playing Agents in Dialogues. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Associ ation for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), 6218 - 6240.h...

    2025 doi

Pith tools

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