Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Large Population Models

T0 review · 4 major / 6 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 Large Population Models claim to simulate millions of agents with realistic, learnable, privacy-preserving behavior by making agent-based models scalable, differentiable, and decentralized.

desk verdict A clear synthesis of the author's prior peer-reviewed ABM work under an LPM umbrella; the headline speedups are asserted without methodology, and Eq. 11 has a fixable but real mathematical error. read the letter →

arxiv 2507.09901 v1 pith:ZVQB3A74 submitted 2025-07-14 cs.MA cs.AI

Ayush Chopra This is my paper
classification cs.MAcs.AI
keywords largepopulationmodelsagent-basedmodelingdifferentiablesimulationLLM-guidedbehaviorarchetypecompressiontensorizedexecutionsecuremulti-partycomputationgradient-basedcalibration
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 introduces Large Population Models (LPMs) as an evolution of agent-based models that can simulate entire populations—millions of agents—with realistic behavior, learn from diverse real-world data streams, and bridge to physical agents while protecting privacy. It claims three innovations together overcome the historic trade-offs of agent-based modeling: a composable, tensorized domain-specific language for scale; end-to-end differentiability for gradient-based calibration and sensitivity analysis; and additive secret sharing for secure decentralized computation. If the claims hold, LPMs would let policymakers test interventions like stimulus checks or vaccine delays in a digital society before deploying them, with applications from pandemic response to supply chains and climate adaptation. The paper anchors the argument with a COVID-19 case study in New York City using 8.4 million agents and reports large computational speedups.

What carries the argument

The machinery is triple. First, FLAME, a domain-specific language that decomposes environment dynamics into modular, differentiable substeps executed as sparse tensor operations on GPUs, exploiting small-world interaction networks and permutation invariance within substeps. Second, the archetype compression scheme: at each time step the simulator identifies K archetypes (demographic and behavioral groups, where K is far smaller than the population N), prompts an LLM M times per archetype and action to estimate a decision distribution, and then each of the N agents samples its action from the archetype's categorical distribution—cutting cost from N queries to K times A times M while keeping heterogeneity through sampling. Third, additive secret sharing over agent states and gradients, which allows each agent to compute aggregate quantities such as infection exposure and derivative terms without learning the private values of any specific neighbor.

What would settle it

Take a mid-sized population with known behavioral heterogeneity, prompt the LLM individually for every agent to obtain fine-grained decision distributions, and compare those against the archetype-sampled version with K small; if the archetype version systematically misses rare but influential subpopulations or produces measurably different emergent system-level curves (for example, epidemic trajectories), the compression claim collapses.

Watch

Extended reading notes

Core claim

The paper's central claim is that the three classic agent-based modeling limitations—scale versus behavioral expressiveness, heterogeneous data assimilation, and the simulation-reality gap—are not inherent but can be jointly solved. It argues that a small number of behavior archetypes queried by a large language model can stand in for millions of individual decision-makers, preserving both inter-group and intra-group heterogeneity through probabilistic sampling; that composing differentiable substeps makes the entire simulation a differentiable computational object, so gradients flow through stochastic events via reparameterization; and that additive secret sharing lets agents compute exposure counts and parameter gradients without revealing neighbors' private states. The NYC case study (8.4 million agents) demonstrates the approach with reported speedups of roughly 200–600x for simulation, 3000x for calibration, and 5000x for analysis, enabling behavior-rich, data-calibrated, privacy-preserving simulation at metropolitan scale.

Load-bearing premise

The load-bearing premise is that the decision-relevant behavior of millions of people can be captured by asking a large language model about a small number of demographic and behavioral archetypes, so that sampling from archetype-level distributions does not lose the heterogeneity that drives real population outcomes.

Editorial extensions

If this is right

  • Epidemiological, economic, and social simulations can run at true metropolitan scale with rich behavioral adaptation, enabling counterfactual policy evaluation (stimulus checks, vaccine schedules, mobility restrictions) in a sandbox before real-world deployment.
  • Gradient-based calibration replaces sample-hungry black-box methods like approximate Bayesian computation, cutting calibration from roughly 100,000 hours to minutes and enabling zero-shot sensitivity analysis from a single forward pass.
  • Privacy-preserving decentralized protocols allow simulations to ingest fresh, granular data from physical agents' devices without centralizing or leaking sensitive information, closing the sim2real loop for adaptive interventions.
  • The same stack transfers beyond health to any domain where individual decisions aggregate into collective outcomes, including supply chain management, disaster response, and financial market design.
  • The AgentTorch open-source framework makes these capabilities practical, so the method can be adopted and stress-tested by other modeling groups.

Reading between the lines

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

  • If the archetype compression holds broadly, the same LLM-prompting trick could serve as a general 'behavior prior' for agent-based models across domains—automating the specification of decision rules that are currently hand-coded.
  • The end-to-end differentiability opens a route beyond calibration: directly optimizing policies (e.g., where to allocate vaccines, which incentives to deploy) by taking gradients through the simulator, although the paper notes that discrete randomness still requires variance reduction.
  • The secret-sharing protocol implies a possible market for private data contributions, where institutions contribute simulation gradients without sharing raw records—but the paper does not quantify the communication and computation overhead of this approach at national scale.
  • A direct and testable extension would compare archetype-derived decision distributions against full agent-level LLM prompting on a held-out population, checking whether emergent system-level outcomes (epidemic curves, unemployment rates) diverge when rare but influential subpopulations matter.
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 paper introduces Large Population Models (LPMs) as an evolution of agent-based models, organized around three pillars: a compositional, tensorized simulation framework (FLAME); end-to-end differentiability for gradient-based calibration and sensitivity analysis; and privacy-preserving decentralized computation via additive secret sharing. It presents a COVID-19 in New York City case study with 8.4 million agents, claims orders-of-magnitude speedups (600x/3000x/5000x in Figure 1, 200x in Section 4.1), and closes with a list of open problems. The manuscript is largely a synthesis of the author's previously published work, with the AgentTorch framework positioned as the implementing system.

Significance. If the stated claims were fully validated, LPMs would be a practically useful framework for large-scale agent-based simulation, combining scale, differentiability, and privacy in a way that is rare in the ABM literature. The paper's strengths include a clear articulation of three real challenges, a concrete open-source implementation (AgentTorch), and a self-critical open-problems section that honestly notes limitations (e.g., Section 7.1, Problem 1). However, the present manuscript does not itself provide a self-contained technical derivation or benchmark evidence for the headline claims; those are deferred to prior papers. The central 'realistic behaviors at population scale' claim rests on an archetype-compression assumption that the paper itself flags as failing for household-level decisions. The paper is better read as a position/vision paper than as a complete research contribution, and the load-bearing empirical and mathematical points need to be fixed or explicitly scoped before the claims can be accepted.

major comments (4)
  1. [Section 4.2, Eqs. (4)-(6); Section 7.1, Problem 1]
  2. [Figure 1; Section 4.1; Figure 10]
  3. [Section 6.1, Eq. (11)]
  4. [Sections 4, 5, and 6 (multiple 'represented in the following papers' passages)]
minor comments (6)
  1. [Section 2.1, Eq. (1)]
  2. [Section 6.1, Eqs. (8)-(9)]
  3. [Section 2.2]
  4. [Section 4.2 and reference [10]]
  5. [Figure 10]
  6. [Section 3.1]

Circularity Check

1 steps flagged · score 6.0 of 10

The archetype method's heterogeneity-preservation claim is self-definitional (Eq. 6 defines the heterogeneity it claims to preserve), while the headline performance benchmarks largely rest on self-citations to prior peer-reviewed work.

  1. self definitional [Section 4.2, Eqs. 4-6, paragraph immediately following Eq. 6]
    "Importantly, this approach preserves both inter-group and intra-group heterogeneity. Even agents belonging to the same archetype may make different decisions due to the probabilistic sampling, while archetypes themselves capture major demographic and behavioral differences."

    The paper presents this as the payoff of Eqs. 4-6, but the 'intra-group heterogeneity' is nothing more than the variance of the Categorical distribution in Eq. 6, and the 'inter-group heterogeneity' is the archetype partition assumed at the start of Section 4.2. No independent observable of real-world decision heterogeneity is defined or measured; the claim that heterogeneity is 'preserved' is equivalent to saying the generative process in Eq. 6 is stochastic and the grouping in Eq. 4 is nontrivial. The conclusion 'without sacrificing behavioral richness' therefore restates the model's construction rather than testing it against data.

full rationale

The paper is largely a position and summary of the author's prior work, so most claims are imported from external peer-reviewed publications rather than derived in-paper. The central archetype method, however, contains a self-definitional step: the heterogeneity that Eq. 6 produces is presented as evidence that the approach 'preserves' behavioral heterogeneity, but that heterogeneity is defined by the very sampling operation being introduced. No external benchmark against individual-level behavior is provided to support the 600x/3000x/5000x claims in Figure 1 and the 200x claim in Section 4.1; those figures are backed by self-citations such as [11], [12], [24], and [25]. Because those cited papers are peer-reviewed and the AgentTorch framework is open-source, the self-citations themselves are not construction-equivalent circularity. The in-scope admission in Section 7.1, Problem 1, that independent archetype sampling 'misses critical correlation structures and leads to unrealistic emergent behaviors' further limits the central 'realistic behaviors' claim to an open problem. Overall, the only genuine circular reduction is the self-definitional heterogeneity-preservation claim; the remaining issues are unsupported or conditional claims rather than circular derivations.

Assumptions & free parameters 4 free parameters · 6 assumptions · 1 invented entities

The central claims lean on the archetype compressibility assumption, sparse interaction structure, reparameterized stochasticity, and an unspecified MPC security model. The paper introduces no new fitted parameters; M, K_t, beta, and n are modeler-chosen or structural quantities. These are not weaknesses per se, but they determine how far the framework's promises extend beyond the demonstrations.

free parameters (4)
  • M (LLM queries per archetype)
    Number of Monte Carlo samples per archetype-action in Eq. 5; modeler-chosen, controls variance and cost of behavioral estimates.
  • K_t (archetype count)
    Number of archetypes at each timestep in Section 4.2; modeler-chosen, determines fidelity-speed tradeoff.
  • beta (effective contact rate)
    Structural parameter in Eq. 8; set during calibration and used in Eq. 11 derivative.
  • n (secret-sharing modulus)
    Modulus in Eqs. 9-10; chosen by the system designer, affects security and overflow behavior.
assumptions (6)
  • domain assumption Small-world interaction networks: each agent interacts with a limited subset of the population, enabling sparse tensor representations.
    Invoked in Section 4.1 as the first principle of tensorized execution; if interaction graphs are dense, the sparse tensorization claim fails.
  • domain assumption Permutation invariance of substeps: order of operations within a substep does not affect the outcome.
    Second principle in Section 4.1; required for tensorized parallel execution to be valid.
  • domain assumption Reparameterization of stochastic mechanisms preserves both distribution and gradient flow.
    Stated in Section 4.1 as the way FLAME maintains differentiability for discrete contact events.
  • domain assumption Archetype compressibility: decision-making processes of millions of agents follow similar patterns by demographic, socioeconomic, and behavioral characteristics, so K_t << N archetypes suffice.
    Core behavioral assumption in Section 4.2, Eqs. 4-6; if false, the scale-versus-expressiveness claim collapses.
  • domain assumption LLM queries provide faithful decision distributions for archetypes.
    Eq. 5 assumes the LLM-based decision function's Monte Carlo samples approximate the true behavior distribution; no validation is provided in this paper.
  • domain assumption Honest-but-curious security model for additive secret sharing.
    Eqs. 9-13 assume semi-honest parties and no collusion to reconstruct shares; the paper does not specify the adversary model, communication channels, or failure modes.
invented entities (1)
  • Dual agent with synthetic and physical existence
    purpose: Bridges simulation and reality so decentralized computation can run where data resides while preserving privacy.
    Introduced in Section 6 as a conceptual notion; no independent measurement or falsifiable prediction beyond the author's prior MPC work is provided in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Large Population Models." pith.science (2026). https://pith.science/paper/ZVQB3A74

@misc{pith2026250709901,
  author       = {Pith},
  title        = {Pith review of: Large Population Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZVQB3A74}},
  note         = {Machine review of arXiv:2507.09901}
}
read the original abstract

Many of society's most pressing challenges, from pandemic response to supply chain disruptions to climate adaptation, emerge from the collective behavior of millions of autonomous agents making decisions over time. Large Population Models (LPMs) offer an approach to understand these complex systems by simulating entire populations with realistic behaviors and interactions at unprecedented scale. LPMs extend traditional modeling approaches through three key innovations: computational methods that efficiently simulate millions of agents simultaneously, mathematical frameworks that learn from diverse real-world data streams, and privacy-preserving communication protocols that bridge virtual and physical environments. This allows researchers to observe how agent behavior aggregates into system-level outcomes and test interventions before real-world implementation. While current AI advances primarily focus on creating "digital humans" with sophisticated individual capabilities, LPMs develop "digital societies" where the richness of interactions reveals emergent phenomena. By bridging individual agent behavior and population-scale dynamics, LPMs offer a complementary path in AI research illuminating collective intelligence and providing testing grounds for policies and social innovations before real-world deployment. We discuss the technical foundations and some open problems here. LPMs are implemented by the AgentTorch framework (github.com/AgentTorch/AgentTorch)

Figures

Figures reproduced from arXiv: 2507.09901 by the authors.

Figure 1
Figure 1. Performance benchmarking comparing computational efficiency of LPMs versus conven [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Trillion-Dollar Pandemic resource allocation breakdown demonstrating the critical role [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Research Pillars: LPMs alleviate the three challenges of ABMs by making simulations [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Domain Specific Language. The simulation has N citizens (Agents) that interact directly and co-locate across pubs (Object) to spread the virus (Object). The simulator state is a collection of properties that describe these entities, is initialized once, and transformed…
Figure 5
Figure 5. Figure 5: Multi-fidelity simulations. (a) Implementation of the FLAME architecture within the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Modeling behavior for 8.4 million agents over 90 steps for $500 (a) LPMs model distinct [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: (a) Differentiable calibration pipeline showing how LPMs transform traditional black-box [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Calibration protocol for a differentiable ABM. Gradient-based learning helps leverage [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Distributed Calibration: We execute calibrate with contextual data distributed across [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Comparative performance analysis between traditional surrogate-based calibration and [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: (a) Secure computation architecture showing how LPMs enable secure aggregation of [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. LLM Economist: Large Population Models and Mechanism Design in Multi-Agent Generative Simulacra

    cs.MA 2025-07 reject novelty 6.0 of 10

    The LLM Economist framework couples persona-conditioned worker agents with an in-context RL planner to search US-bracket tax schedules, yet its Saez benchmark is derived from the planner's own solution and its headlin...

Reference graph

Works this paper leans on

27 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [1]

    Reuters, August 2021

    Indonesia probes suspected data breach on COVID-19 app. Reuters, August 2021

    work page 2021

  2. [2]

    Google COVID-19 community mobility reports: Anonymization process description (version 1.1)

    Ahmet Aktay, Shailesh Bavadekar, Gwen Cossoul, John Davis, Damien Desfontaines, Alex Fabrikant, Evgeniy Gabrilovich, Krishna Gadepalli, Bryant Gipson, Miguel Guevara, et al. Google COVID-19 community mobility reports: Anonymization process description (version 1.1). arXiv preprint arXiv:2004.04145, 2020

    arXiv 2004

  3. [3]

    Robert L. Axtell. 120 Million Agents Self-Organize into 6 Million Firms: A Model of the U.S. Private Sector. In Proceedings of the 2016 International Conference on Autonomous Agents & Multiagent Systems, AAMAS ’16, pages 806–816, Richland, SC, May 2016. International Foundation for Autonomous Agents and Multiagent Systems

    work page 2016

  4. [4]

    Axtell and J

    Robert L. Axtell and J. Doyne Farmer. Agent-Based Modeling in Economics and Finance: Past, Present, and Future. Journal of Economic Literature

    work page

  5. [5]

    June: open-source individual-based epidemiology simulation

    Joseph Aylett-Bullock, Carolina Cuesta-Lazaro, Arnau Quera-Bofarull, Miguel Icaza-Lizaola, Aidan Sedgewick, Henry Truong, Aoife Curran, Edward Elliott, Tristan Caulfield, Kevin Fong, et al. June: open-source individual-based epidemiology simulation. Royal Society open science, 8(7):210506, 2021

    work page 2021

  6. [6]

    June: Open-source individual-based epidemiology simulation

    Joseph Aylett-Bullock, Carolina Cuesta-Lazaro, Arnau Quera-Bofarull, Miguel Icaza-Lizaola, Aidan Sedgewick, Henry Truong, Aoife Curran, Edward Elliott, Tristan Caulfield, Kevin Fong, Ian Vernon, Julian Williams, Richard Bower, and Frank Krauss. June: Open-source individual-based epidemiology simulation. Royal Society Open Science , 8(7):210506, July 2021

    work page 2021

  7. [7]

    Epidemiological modelling in refugee and internally displaced people settlements: challenges and ways forward

    Joseph Aylett-Bullock, Robert Tucker Gilman, Ian Hall, David Kennedy, Egmond Samir Evers, Anjali Katta, Hussien Ahmed, Kevin Fong, Keyrellous Adib, Lubna Al Ariqi, et al. Epidemiological modelling in refugee and internally displaced people settlements: challenges and ways forward. BMJ Global Health, 7(3):e007822, 2022

    work page 2022

  8. [8]

    Heterogeneous effects and spillovers of macroprudential policy in an agent-based model of the UK housing market

    Adrian Carro, Marc Hinterschweiger, Arzu Uluc, and J Doyne Farmer. Heterogeneous effects and spillovers of macroprudential policy in an agent-based model of the UK housing market. Industrial and Corporate Change, 32(2):386–432, April 2023

    work page 2023

Show all 27 references
  1. [9]

    Generation of Synthetic Populations in Social Simulations: A Review of Methods and Practices

    Kevin Chapuis, Patrick Taillandier, and Alexis Drogoul. Generation of Synthetic Populations in Social Simulations: A Review of Methods and Practices. Journal of Artificial Societies and Social Simulation, 25(2):6, 2022. 18

    2022

  2. [10]

    A framework for learning in agent-based models

    Ayush Chopra, Shashank Kumar, Nurullah Giray-Kuru, Ramesh Raskar, and Arnau Quera- Bofarull. A framework for learning in agent-based models. In Proceedings of the 24th Interna- tional Conference on Autonomous Agents and Multi-Agent Systems. International Foundation for Autonom...

    2025

  3. [11]

    Aditya Prakash, and Ramesh Raskar

    Ayush Chopra, Alexander Rodríguez, Jayakumar Subramanian, Arnau Quera-Bofarull, Balaji Krishnamurthy, B. Aditya Prakash, and Ramesh Raskar. Differentiable Agent-based Epi- demiology. In Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent System...

    2023

  4. [12]

    A frame- work for learning in agent-based models

    Ayush Chopra, Jayakumar Subramanian, Balaji Krishnamurthy, and Ramesh Raskar. A frame- work for learning in agent-based models. In Proceedings of the 23rd International Conference on Autonomous Agents and Multi-Agent Systems. International Foundation for Autonomous Agents and ...

    2024

  5. [13]

    T-Mobile ‘Put My Life in Danger’ Says Woman Stalked With Black Market Location Data, August 2019

    Joseph Cox. T-Mobile ‘Put My Life in Danger’ Says Woman Stalked With Black Market Location Data, August 2019

    2019

  6. [14]

    Doyne Farmer, and Sebastian M

    Joel Dyer, Patrick Cannon, J. Doyne Farmer, and Sebastian M. Schmon. Black-box Bayesian inference for agent-based models. Journal of Economic Dynamics and Control, 161:104827, April 2024

    2024

  7. [15]

    Acts of congress and covid-19: A literature review on the impact of increased unemployment insurance benefits and stimulus checks

    Elena Falcettoni and Vegard M Nygaard. Acts of congress and covid-19: A literature review on the impact of increased unemployment insurance benefits and stimulus checks. 2020

    2020

  8. [16]

    Agent-based modelling of post-disaster recovery with remote sensing data

    Saman Ghaffarian, Debraj Roy, Tatiana Filatova, and Norman Kerle. Agent-based modelling of post-disaster recovery with remote sensing data. International Journal of Disaster Risk Reduction, 60:102285, 2021

    2021

  9. [17]

    Robert Hinch, William J. M. Probert, Anel Nurtay, Michelle Kendall, Chris Wymant, Matthew Hall, Katrina Lythgoe, Ana Bulas Cruz, Lele Zhao, Andrea Stewart, Luca Ferretti, Daniel Montero, James Warren, Nicole Mather, Matthew Abueg, Neo Wu, Olivier Legat, Katie Bentley, Thomas M...

    2021

  10. [18]

    Illinois Bought Invasive Phone Location Data From Banned Broker Safegraph

    Bennett Cyphers and Jason Kelley. Illinois Bought Invasive Phone Location Data From Banned Broker Safegraph. https://www.eff.org/deeplinks/2021/08/illinois-bought-invasive- phone-location-data-banned-broker-safegraph, August 2021

    2021

  11. [19]

    Kerr, Robyn M

    Cliff C. Kerr, Robyn M. Stuart, Dina Mistry, Romesh G. Abeysuriya, Katherine Rosenfeld, Gregory R. Hart, Rafael C. Núñez, Jamie A. Cohen, Prashanth Selvaraj, Brittany Hagedorn, Lauren George, Michał Jastrz˛ ebski, Amanda S. Izzo, Greer Fowler, Anna Palmer, Dominic Delport, Nic...

    2021

  12. [20]

    Impacts of covid-19 lockdowns and stimulus payments on low-income population’s spending in the united states

    Kangli Li, Natasha Zhang Foutz, Yuxin Cai, Yunlei Liang, and Song Gao. Impacts of covid-19 lockdowns and stimulus payments on low-income population’s spending in the united states. PloS one, 16(9):e0256407, 2021

    2021

  13. [21]

    Robert E. Lucas. Econometric policy evaluation: A critique. Carnegie-Rochester Conference Series on Public Policy, 1:19–46, 1976

    1976

  14. [22]

    Martin, David R

    Christopher A. Martin, David R. Jenkins, Jatinder S. Minhas, Laura J. Gray, Julian Tang, Caroline Williams, Shirley Sze, Daniel Pan, William Jones, Raman Verma, Scott Knapp, Rupert Major, Melanie Davies, Nigel Brunskill, Martin Wiselka, Chris Brightling, Kamlesh Khunti, Pranab...

    2020

  15. [23]

    O’Brien, Carrie J

    Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative Agents: Interactive Simulacra of Human Behavior, August 2023

    2023

  16. [24]

    Don’t simulate twice: one-shot sensitivity analyses via automatic differentiation

    Arnau Quera-Bofarull, Ayush Chopra, Joseph Aylett-Bullock, Carolina Cuesta-Lazaro, Anisoara Calinescu, Ramesh Raskar, and Michael Wooldridge. Don’t simulate twice: one-shot sensitivity analyses via automatic differentiation. 2023

    2023

  17. [25]

    Public health impact of delaying second dose of bnt162b2 or mrna-1273 covid-19 vaccine: simulation agent based modeling study

    Santiago Romero-Brufau, Ayush Chopra, Alex J Ryu, Esma Gel, Ramesh Raskar, Walter Kremers, Karen S Anderson, Jayakumar Subramanian, Balaji Krishnamurthy, Abhishek Singh, et al. Public health impact of delaying second dose of bnt162b2 or mrna-1273 covid-19 vaccine: simulation a...

    2021

  18. [26]

    Covid-19 is linked to changes in the time–space dimension of human mobility

    Clodomir Santana, Federico Botta, Hugo Barbosa, Filippo Privitera, Ronaldo Menezes, and Riccardo Di Clemente. Covid-19 is linked to changes in the time–space dimension of human mobility. Nature Human Behaviour, 7(10):1729–1739, 2023

    2023

  19. [27]

    Agapiou, Avia Aharon, Ron Ziv, Jayd Matyas, Edgar A

    Alexander Sasha Vezhnevets, John P. Agapiou, Avia Aharon, Ron Ziv, Jayd Matyas, Edgar A. Duéñez-Guzmán, William A. Cunningham, Simon Osindero, Danny Karmon, and Joel Z. Leibo. Generative agent-based modeling with actions grounded in physical, social, or digital space using Con...

    2023

Pith tools

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