Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Llama-3.1-8B reproduces human cooperation patterns in 121 dyadic games more accurately than the Nash equilibrium does, indicating that LLMs can serve as digital twins for behavioral experiments.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 23:38 UTC pith:LQQJDZVF

load-bearing objection Tuned-on-the-target pipeline weakens the headline replication, but the systematic grid and preregistered extension merit a serious referee. the 4 major comments →

arxiv 2511.04500 v3 pith:LQQJDZVF submitted 2025-11-06 cs.AI cs.CLcs.GTcs.MA

Large language models replicate and predict human cooperation across experiments in game theory

Andrea Cera Palatsi , Samuel Martin-Gutierrez , Ana S. Cardenal , Max Pellert This is my paper
classification cs.AI cs.CLcs.GTcs.MA MSC 91A1091A2291A90
keywords large language modelsgame theorycooperationNash equilibriumbehavioral phenotypesdigital twinsprompt engineeringexperimental economics
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that a carefully prompted open-weight language model can reproduce aggregate human cooperation in classic two-player games, and can do so more faithfully than the standard rational-choice benchmark. Using 121 games spanning four classical game types, the authors show that Llama's average cooperation rates closely track human data (MSD 0.031, r=0.89), while the Nash equilibrium deviates more (MSD 0.096, r=0.78). A different model, Qwen, aligns with Nash rather than with humans, showing that model choice changes the behavioral phenotype. The same setup is then extended to 320 novel game configurations to generate preregistered predictions for future human experiments. If the claim holds, behavioral scientists could explore experimental parameter spaces in silico before committing to human studies.

Core claim

The central claim is that Llama-3.1-8B, prompted in neutral language without game-theory jargon and filtered through a multi-step reasoning-and-verification pipeline, yields cooperation rates across 121 (S,T) payoff combinations that replicate empirical human patterns with high accuracy. The key signature is high cooperation where the sucker payoff S equals or exceeds the temptation payoff T, mirroring the 'envious' decision rule dominant in human data, and low cooperation where T exceeds the reward R. Llama reproduces this pattern better than Nash equilibrium does, which the paper presents as evidence that LLMs can act as digital twins: first replicating existing experiments, then generatin

What carries the argument

The load-bearing mechanism is the prompt-and-extraction pipeline: instructions use neutral vocabulary (A/B choices, 'other player'), state a one-shot simultaneous game with direct euro-per-point payment, and ask the model to reason step by step; responses are then classified as good or bad by a second LLM acting as a logical verifier before the chosen label is extracted. Progressively adding these layers transforms Llama's nearly random one-word answers into a structured cooperation matrix. The comparisons use mean squared displacement and Pearson correlation between the model, human, and Nash cooperation matrices, with Nash equilibria computed via replicator dynamics for each game.

Load-bearing premise

The claim rests on the assumption that the adapted instructions and the Qwen verifier measure the same decision problem humans faced, rather than changing the game (for example, making point payoffs more salient than the original lottery mechanism) and thereby manufacturing the apparent match.

What would settle it

Run the preregistered human experiment on the extended S/T grid: if humans do not show the S≥T high-cooperation diagonal and the T=R boundary that Llama predicts, the extrapolation claim is refuted. Alternatively, turn off the logical verifier and recompute the MSD of Llama against human data; if the match degrades sharply, the fidelity is produced by the filter rather than by the model's underlying decision process.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Llama reproduces population-level human cooperation without persona-based prompting, so building synthetic participant pools can be simpler than earlier approaches.
  • Different models embody different decision phenotypes: Qwen behaves like a rational (Nash) player, which makes some LLMs suitable human simulators and others not.
  • The paper's predicted cooperation rates for novel parameter combinations are preregistered, giving a direct way to test whether LLM or Nash extrapolation better anticipates real human behavior.
  • Rational-choice predictions are less accurate than this LLM at capturing human deviations, so LLM-based simulation may complement formal game theory as a descriptive model.

Where Pith is reading between the lines

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

  • The reported fidelity is conditional on the verifier: up to 25% of responses in some games bypassed it, and without it Llama's matrix degrades, so an ablation removing the verifier would show how much of the human-like pattern is a measurement effect rather than model behavior.
  • If the preregistered human experiments confirm Llama's novel-game predictions, LLM-based simulation would be a credible low-cost instrument for hypothesis generation; if they fail, the method is better read as a tool that recovers known experiments, not one that predicts new ones.
  • The envious profile shared by humans and Llama may reflect a payoff-comparison heuristic absorbed from text rather than genuine social preference; a scale-invariance test—varying whether the prize is framed as euros or lottery tickets—could distinguish those accounts.
  • The random-like cooperation near (S,T)=(0,0) could be a salience artifact of near-zero payoffs; a targeted simulation with small nonzero payoffs would test whether this boundary is robust or an edge case of numeric framing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper claims that an 8B open-weight LLM (Llama-3.1-8B), when equipped with a specific multi-step reasoning prompt and a logical verifier, replicates aggregate human cooperation patterns in 121 dyadic 2×2 games better than the Nash equilibrium does (Table 2: MSD=0.031, r=0.89 vs. MSD=0.096, r=0.78). The authors compare three open models, characterize behavioral phenotypes, and use Llama to extrapolate to novel games, with preregistered hypotheses for future human experiments. The central claim is that the extraction pipeline is a neutral measurement of LLM behavior, revealing an emergent human-like pattern rather than constructing it.

Significance. If established with a confirmatory analysis, this result would be a valuable contribution: it suggests that an open, reproducible LLM-based pipeline can serve as a 'digital twin' for aggregate human behavior in simple strategic interactions, enabling low-cost in-silico hypothesis generation. The paper's strengths include the use of open-source models, public code and data, preregistration of the extended experiments, and transparent reporting of limitations (e.g., verifier bypass). However, as written, the central quantitative claim is compromised by the fact that the prompt and verifier were explicitly selected to maximize cooperation in the S≥T region, which is exactly the region where human cooperation is high and where Llama's matrix shows its most distinctive pattern. Thus the headline fidelity metrics in Table 2 are in-sample fits, not out-of-sample replication.

major comments (4)
  1. [§3.4.3, §3.4.4, Table 2] The multi-step prompt and logical verifier were selected to maximize cooperation in the S≥T region (Harmony Games), as stated in §3.4.3 and §3.4.4. This is precisely the region where human cooperation is highest (Fig. 1, far left) and where the final Llama matrix shows its most distinctive pattern. The near-random results under Simple Extraction (Fig. 1) show that the measured high fidelity is not a default property of Llama but is produced by the tuned layers. Consequently, the MSD=0.031, r=0.89 values in Table 2 measure agreement between a pipeline tuned to the target human matrix and the human matrix itself; they cannot be read as out-of-sample replication. The authors should pre-register the selection rule before inspecting human data, conduct a robustness analysis over a family of prompt/verifier variants, or hold out a subset of games for validation.
  2. [Table 2] No confidence intervals or significance tests are reported for MSD or Pearson's r. Each cooperation-rate cell is the mean of 20 stochastic samples at temperature 0.8; the standard error of a proportion near 0.5 is approximately 0.11. The human matrix is also a finite sample from ~500 participants, yet it is treated as a fixed benchmark. Bootstrap or other resampling procedures are needed to assess whether the reported differences (e.g., Llama vs. Nash: MSD 0.031 vs. 0.096, r 0.89 vs. 0.78) are statistically robust or could be within sampling noise. Without this, the quantitative ranking is not fully supported.
  3. [§3.3] The original human experiment paid participants with lottery tickets according to points scored, while the LLM prompt replaces this with a direct monetary conversion ('10 euros per point' with a worked example). This changes the decision problem from a lottery (involving risk preferences) to a certain monetary payoff. Cooperative and defective choices can be sensitive to risk attitudes, so the LLM may be solving a different task from the human participants. The paper should justify this deviation or, ideally, rerun the simulation with a lottery-based or otherwise matched incentive structure to show the qualitative pattern is robust to this change.
  4. [§3.6] The 'Nash equilibrium' matrix is computed using replicator dynamics with a specific equilibrium selection: in the bistable Stag Hunt region, the initial condition x(0)=0.5 determines the boundary between the basins of attraction. Nash equilibria are not unique in coordination games; other standard selections (payoff-dominant, risk-dominant, or mixed) would yield different cooperation rates. The claim that Llama is closer to human data than Nash is therefore benchmark-dependent. The authors should either justify this equilibrium selection as the appropriate theoretical benchmark or report results under alternative equilibrium concepts.
minor comments (4)
  1. [Fig. 1 and Fig. A1 captions] The color scale descriptions are inconsistent: Fig. 1 says '0 (purple: no cooperation) to 1 (yellow: full cooperation)', while Fig. A1 says '0 (yellow: no cooperation) to 1 (purple: full cooperation)'. Please use a single convention and ensure the captions match the actual colormaps.
  2. [Table 3] The extraction accuracy rates (0.97, 0.96, 0.83) are presented without confidence intervals or the number of samples per model beyond the statement of 100 long answers total. Reporting the per-model sample sizes and a measure of uncertainty would strengthen the comparison.
  3. [§A.4] Bypass rates reach 0.25 for some games, meaning a quarter of responses in those cells did not pass the logical verifier. The paper states the impact 'remains limited', but a sensitivity analysis excluding or flagging bypassed responses would quantitatively support that assertion and is directly relevant to the noise in Table 2.
  4. [§1.2] The reference to 'non-transparent regions of Figure 2' is somewhat confusing because the figure uses transparency only in panel A. Clarify the wording or ensure the figure's legend makes the delineation explicit.

Circularity Check

3 steps flagged

The headline replication is partly fitted: prompt and verifier were selected to maximize cooperation in the S≥T region, and the reported 'envious' phenotype is defined by that same condition.

specific steps
  1. fitted input called prediction [§3.4.3 (Multi-step prompt); used in §1.2/Table 2]
    "To identify the optimal prompt, we evaluated performance in the S≥T region (Harmony Games), where cooperation is the only sensible choice. We selected the prompt version that yielded the highest average cooperation rates for both Llama and Mistral in this region."

    The paper's central claim (Llama 'reproduces human cooperation patterns with high fidelity', MSD=0.031, r=0.89) is scored against the human matrix that has 'a clear area defined by S≥T that displays distinctly high cooperation'. The prompt was explicitly tuned to maximize cooperation in exactly this S≥T region, so the human-like S≥T diagonal in the final Llama matrix is partly created by the selection step. The Simple Extraction baseline (Fig. 1) shows default Llama cooperation is 'almost random'; the human-like pattern appears only after tuning.

  2. fitted input called prediction [§3.4.4 (Logical verifier); affects Figure 1 and Table 2]
    "Following the same validation approach used for the multi-step prompt, we selected the version that maximized cooperation rates in the S≥T region (Harmony Games)."

    The logical verifier is a second selection layer optimized on the same S≥T objective. It filters the long-form responses before extraction, so the final cooperation matrix entering Table 2 is twice biased toward the same region that dominates the human–Llama similarity. The near-random result under Simple Extraction shows this is not a default Llama property but a product of the tuned extraction pipeline.

  3. self definitional [Phenotype list (intro) + Abstract + §3.4.3/§3.4.4]
    "Envious. Seeks to maximize relative advantage over the opponent. Cooperates if S≥T."

    The abstract reports that 'humans and Llama share an envious decision profile', but the Envious phenotype is defined as 'Cooperates if S≥T' and the prompt/verifier were selected to maximize cooperation in precisely the S≥T region. Thus the envious profile is substantially installed by the selection criterion; finding it in Llama is not an independent behavioral discovery, although the exact cooperation rates and other regions (e.g., T>R) are not fully determined by this criterion.

full rationale

The paper's genuinely out-of-sample contribution is the extended-grid Llama simulation and preregistration (§1.3, Discussion), and the Nash-equilibrium comparison is a fixed external benchmark rather than a fitted input; these reduce the overall circularity. However, the original-grid replication headline in §1.2/Table 2 is not a neutral measurement: the S≥T region is simultaneously (a) a distinctive human high-cooperation region, (b) the explicit optimization target for both the multi-step prompt (§3.4.3) and logical verifier (§3.4.4), and (c) the defining condition of the Envious phenotype claimed in the abstract. The multi-step criterion has an independent normative rationale (cooperation is the only sensible choice in Harmony games), so the circularity is partial rather than total; default Llama behavior is 'almost random' (Fig. 1), and the human-like pattern emerges only after the tuned layers. No load-bearing self-citations or imported uniqueness theorems were found. Score 6 reflects one or more construction-based predictions with partial independent content remaining.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central result leans on a pipeline of modeling choices (prompt wording, multi-step decomposition, verifier filtering) rather than on a fixed, externally specified procedure. Two of those choices were tuned against the human cooperation pattern in the S≥T region. The Nash benchmark derivation itself is standard and self-contained.

free parameters (3)
  • Multi-step prompt variant selection = variant maximizing cooperation in S≥T (Harmony) region
    §3.4.3: 'We selected the prompt version that yielded the highest average cooperation rates for both Llama and Mistral in this region.' This tunes the pipeline toward the human pattern being replicated.
  • Logical verifier prompt/example selection = version maximizing cooperation in S≥T region
    §3.4.4: 'Following the same validation approach... we selected the version that maximized cooperation rates in the S≥T region (Harmony Games).'
  • Verifier bypass threshold = disable after one round with no accepted games
    §3.5 adaptive relaxation; affects up to 25% of responses per game (Fig. A2), not corrected in analysis.
axioms (4)
  • domain assumption Human comparison data from Poncela-Casasnovas et al. (2016) are accurate and directly comparable to LLM play
    All replication metrics in Table 2 use this dataset as ground truth; any errors or framing differences propagate into the conclusion.
  • domain assumption LLM responses at temperature 0.8 over 20 repetitions represent a stable behavioral distribution
    Cooperation rates are computed from 20 stochastic samples; no variance or confidence interval is reported.
  • domain assumption The Qwen logical verifier correctly classifies valid vs. invalid reasoning and does not bias cooperation rates
    Verifier accuracy is not measured; only extraction accuracy is; up to 25% of responses bypass verification (Fig. A2).
  • standard math Replicator dynamics and fixed-point analysis correctly characterize Nash equilibria of 2x2 symmetric games
    §3.6; used as the rational benchmark against which LLM behavior is compared.

pith-pipeline@v1.3.0-alltime-deepseek · 19730 in / 11091 out tokens · 107534 ms · 2026-08-03T23:38:19.700858+00:00 · methodology

0 comments
read the original abstract

Large language models (LLMs) are increasingly deployed as decision-making agents in high-stakes domains and as imitators of human behavior in the social and behavioral sciences. Yet how closely LLMs mirror human decision-making remains poorly understood. This gap is critical: misalignment could produce harmful outcomes in practice, while failure to replicate human behavior renders LLMs ineffective as social simulators. Here, we address this gap by replicating large-scale game-theoretic experiments and by introducing a systematic prompting and probing framework for machine-behavioral evaluation. We test three open models typically used to power agents (Llama, Mistral, and Qwen). Across 121 dyadic games spanning four classical game types, Llama reproduces human cooperation patterns with high fidelity, while Qwen aligns closely with Nash equilibrium predictions. Characterizing models through behavioral phenotyping, we find that humans and Llama share an envious decision profile, while Qwen and Mistral exhibit different profiles. An attention-based analysis of payoff salience reveals Llama processes payoff information in a structured, layer-dependent manner absent in Qwen and Mistral, suggesting a mechanistic basis for its closer alignment with human behavior. Population-level behavioral replication is achieved without persona-based prompting, simplifying the simulation process. Extending the experimental parameter space beyond the original human-tested games, we generate and preregister testable hypotheses for novel game configurations. Our findings demonstrate appropriately configured LLMs can replicate aggregate human behavioral patterns, exhibit human-like decision phenotypes, and enable systematic exploration of unexplored experimental spaces, offering a complementary approach to traditional behavioral research that generates new empirical predictions about human social decision-making.

Figures

Figures reproduced from arXiv: 2511.04500 by Ana S. Cardenal, Andrea Cera Palatsi, Max Pellert, Samuel Martin-Gutierrez.

Figure 1
Figure 1. Figure 1: Cooperation patterns of humans and LLMs. Left panel: Human cooperation matrix from the original experiments22 with an average cooperation of 0.480. Each quadrant of the parameter grid corresponds to a standard game in game theory: HG (Harmony Game), SG (Snowdrift Game), SH (Stag Hunt Game), and PD (Prisoner’s Dilemma). Games outside these quadrants represent non-standard game configurations. Central panel:… view at source ↗
Figure 2
Figure 2. Figure 2: Cooperation matrices of Llama and the Nash equilibrium, extended through simulations. Each tile represents a game defined by the parameter combination of (S,T), with average cooperation ranging from 0 (purple: no cooperation) to 1 (yellow: full cooperation). Panel A: Cooperation matrix generated by Llama simulations. The non-transparent region delineated by a black border shows simulations with parameter c… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

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

  1. Strategy, Not Payoffs: A Behavioural Embedding of Normal-Form Games

    cs.GT 2026-07 conditional novelty 6.0

    A two-feature game embedding (Nash entropy and best-response switching) predicts cross-game transfer of fine-tuned LLMs on held-out games, outperforming game identity and published structural embeddings.

  2. Large language models converge on competitive rationality but diverge on cooperation across providers and generations

    physics.soc-ph 2026-04 unverdicted novelty 6.0

    LLMs converge on competitive rationality and coordination but diverge 48-fold on cooperation, with provider identity and generational shifts as dominant factors across 38 games.

Reference graph

Works this paper leans on

39 extracted references · 6 canonical work pages · cited by 2 Pith papers · 1 internal anchor

  1. [1]

    P.et al.Out of One, Many: Using Language Models to Simulate Human Samples.Polit

    Argyle, L. P.et al.Out of One, Many: Using Language Models to Simulate Human Samples.Polit. Analysis31, 1–15, DOI: 10.1017/pan.2023.2 (2023). 2.Rahwan, I.et al.Machine behaviour.Nature568, 477–486, DOI: 10.1038/s41586-019-1138-y (2019)

    doi:10.1017/pan.2023.2 2023

  2. [2]

    Group the outcomes based on your decision: - If you choose A: (A,A) and (A,B) - If you choose B: (B,A) and (B,B)

  3. [3]

    Meng, X.et al.The application of large language models in medicine: A scoping review.iScience27, 109713–109713, DOI: 10.1016/j.isci.2024.109713 (2024)

    arXiv 2024

  4. [4]

    J.et al.Large language models in medicine.Nat

    Thirunavukarasu, A. J.et al.Large language models in medicine.Nat. Med.29, 1930–1940, DOI: 10.1038/ s41591-023-02448-8 (2023). 5.Lai, J.et al.Large language models in law: A survey (2023). 2312.03718

    Pith/arXiv arXiv 1930

  5. [5]

    18/24 Logical verifier prompt

    Based on your analysis, choose either A or B. 18/24 Logical verifier prompt. You have to read the answer of Player 1 and output good or bad according to the following criteria: Criteria for good: - ALL arithmetic/mathematic comparisons and claims are CORRECT. - ALL descriptions of the possible outcomes are CORRECT. - ONLY TWO possible outputs per choice, ...

  6. [6]

    & Z˝odi, Z

    Homoki, P. & Z˝odi, Z. Large language models and their possible uses in law.Hungarian J. Leg. Stud.64, 435–455, DOI: 10.1556/2052.2023.00475 (2024). 7.Wang, S.et al.Large language models for education: A survey and outlook (2024). 2403.02334. 8.Grassucci, E.et al.Beyond answers: How llms can pursue strategic thinking in education (2025). 2504.04815. 9.Pas...

    arXiv 2052

  7. [10]

    Binz, M.et al.A foundation model to predict and capture human cognition.Nature644, 1002–1009, DOI: 10.1038/ s41586-025-09215-4 (2025)

    2025

  8. [11]

    & Zhu, J.-Q

    Xie, H. & Zhu, J.-Q. Centaur May Have Learned a Shortcut that Explains Away Psychological Tasks, DOI: 10.31234/osf. io/u7z4t_v1 (2025)

    doi:10.31234/osf 2025

  9. [12]

    Sturm, B. L. A Simple Method to Determine if a Music Information Retrieval System is a “Horse”.IEEE Transactions on Multimed.16, 1636–1644, DOI: 10.1109/TMM.2014.2330697 (2014)

    arXiv 2014

  10. [13]

    Horton, J. J. Large language models as simulated economic agents: What can we learn from homo silicus? (2023). 2301.07543

    arXiv 2023

  11. [14]

    Aher, G., Arriaga, R. I. & Kalai, A. T. Using large language models to simulate multiple humans and replicate human subject studies. InProceedings of the International Conference on Machine Learning (ICML), DOI: 10.48550/arXiv.2208.10264 (2023)

    Pith/arXiv arXiv doi:10.48550/arxiv.2208.10264 2023

  12. [15]

    & Jackson, M

    Mei, Q., Xie, Y ., Yuan, W. & Jackson, M. O. A turing test of whether ai chatbots are behaviorally similar to humans.Proc. Natl. Acad. Sci. USA121, DOI: 10.1073/pnas.2313925121 (2024). 16.Hagendorff, T.et al.Machine Psychology, DOI: 10.48550/arXiv.2303.13988 (2024). 2303.13988

    Pith/arXiv arXiv doi:10.48550/arxiv.2303.13988 2024

  13. [17]

    M., Wagner, C., Rammstedt, B

    Pellert, M., Lechner, C. M., Wagner, C., Rammstedt, B. & Strohmaier, M. AI Psychometrics: Assessing the Psychological Profiles of Large Language Models Through Psychometric Inventories.Perspectives on Psychol. Sci.DOI: 10.1177/ 17456916231214460 (2024). 18.Grattafiori, A.et al.The Llama 3 Herd of Models, DOI: 10.48550/arXiv.2407.21783 (2024). 2407.21783. ...

    Pith/arXiv arXiv doi:10.48550/arxiv.2407.21783 2024

  14. [22]

    Poncela-Casasnovas, J.et al.Humans display a reduced set of consistent behavioral phenotypes in dyadic games.Sci. Adv. 2, e1600451, DOI: 10.1126/sciadv.1600451 (2016). 23.Evans, J. & Duede, E. After science.Science390, eaec7650, DOI: 10.1126/science.aec7650 (2025)

    work page doi:10.1126/sciadv.1600451 2016

  15. [24]

    D., Tsurusaki, Y ., Fukasawa, Y

    Kitadai, A., Rico Lugo, S. D., Tsurusaki, Y ., Fukasawa, Y . & Nishino, N. Can AI with High Reasoning Ability Replicate Human-like Decision Making in Economic Experiments?Group Decis. Negot.34, 1303–1326, DOI: 10.1007/ s10726-025-09946-9 (2025)

    2025

  16. [25]

    Simon, H. A. A Behavioral Model of Rational Choice.The Q. J. Econ.69, 99–118, DOI: 10.2307/1884852 (1955). 1884852

    doi:10.2307/1884852 1955

  17. [26]

    The simplicity principle in perception and cognition.WIREs Cogn

    Feldman, J. The simplicity principle in perception and cognition.WIREs Cogn. Sci.7, 330–340, DOI: 10.1002/wcs.1406 (2016)

    work page doi:10.1002/wcs.1406 2016

  18. [27]

    Martin-Gutierrez, S., van Dissel, M. N. C. & Karimi, F. The hidden architecture of connections: How do multidimensional identities shape our social networks?, DOI: 10.48550/arXiv.2406.17043 (2024). 2406.17043

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2406.17043 2024

  19. [28]

    A., Dunbar, R

    Tamarit, I., Cuesta, J. A., Dunbar, R. I. M. & Sánchez, A. Cognitive resource allocation determines the organization of personal networks.Proc. Natl. Acad. Sci.115, 8316–8321, DOI: 10.1073/pnas.1719233115 (2018)

    work page doi:10.1073/pnas.1719233115 2018

  20. [29]

    M., Gebru, T., McMillan-Major, A

    Bender, E. M., Gebru, T., McMillan-Major, A. & Shmitchell, S. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? InProceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–623, DOI: 10.1145/3442188.3445922 (ACM, Virtual Event Canada, 2021)

    arXiv 2021

  21. [30]

    & Törnberg, P

    Barrie, C. & Törnberg, P. Emergent LLM behaviors are observationally equivalent to data leakage, DOI: 10.48550/arXiv. 2505.23796 (2025). 2505.23796

    Pith/arXiv arXiv doi:10.48550/arxiv.2505.23796 2025

  22. [31]

    & Chilton, L

    Sreedhar, K. & Chilton, L. Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs, DOI: 10.48550/arXiv.2402.08189 (2024). 2402.08189. 32.Wei, J.et al.Chain-of-thought prompting elicits reasoning in large language models (2022). 2201.11903

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.08189 2024

  23. [32]

    There are four possible outcomes in this game, depending on your choice (A or B) and the other player’s choice (A or B)

  24. [33]

    & Törnberg, P

    Larooij, M. & Törnberg, P. Do Large Language Models Solve the Problems of Agent-Based Modeling? A Critical Review of Generative Social Simulations, DOI: 10.48550/arXiv.2504.03274 (2025). 2504.03274

    Pith/arXiv arXiv doi:10.48550/arxiv.2504.03274 2025

  25. [34]

    Fagbohun, O., Harrison, R. M. & Dereventsov, A. An empirical categorization of prompting techniques for large language models: A practitioner’s guide.J. Artif. Intell. Mach. Learn. Data Sci.1, 1–11, DOI: 10.51219/jaimld/oluwole-fagbohun/15 (2023). 35.Amatriain, X. Prompt design and engineering: Introduction and advanced methods (2024). 2401.14423. 36.Koji...

    Pith/arXiv arXiv 2023

  26. [35]

    Analyze and compare the outcomes in each group step by step

  27. [36]

    Make sure your analysis is consistent with the rules of the game and the possible outcomes

  28. [37]

    & Strohmaier, M

    Ahnert, G., Haensch, A.-C., Plank, B. & Strohmaier, M. Survey Response Generation: Generating Closed-Ended Survey Responses In-Silico with Large Language Models, DOI: 10.48550/arXiv.2510.11586 (2025). 2510.11586

    Pith/arXiv arXiv doi:10.48550/arxiv.2510.11586 2025

  29. [38]

    B.et al.Language models are few-shot learners

    Brown, T. B.et al.Language models are few-shot learners. InAdvances in Neural Information Processing Systems, vol. 33, 1877–1901, DOI: 10.48550/arXiv.2005.14165 (2020). 39.Skyrms, B. The Stag Hunt.Proc. Addresses Am. Philos. Assoc.75, 31, DOI: 10.2307/3218711 (2001). 3218711

    Pith/arXiv arXiv doi:10.48550/arxiv.2005.14165 1901

  30. [39]

    - (A,B): You earn 10 points, and the other player earns 6 points

    If you choose A: - (A,A): Both players earn 10 points each. - (A,B): You earn 10 points, and the other player earns 6 points

  31. [40]

    & Chammah, A

    Rapoport, A. & Chammah, A. M. The Game of Chicken.Am. Behav. Sci.10, 10–28, DOI: 10.1177/000276426601000303 (1966). 41.Rapoport, A. & Chammah, A. M.Prisoner’s Dilemma(University of Michigan Press, 1965)

    work page doi:10.1177/000276426601000303 1966

  32. [41]

    - (B,B): Both players earn 5 points each

    If you choose B: - (B,A): You earn 6 points, and the other player earns 10 points. - (B,B): Both players earn 5 points each

  33. [42]

    Kahneman, D., Knetsch, J. L. & Thaler, R. H. Fairness as a constraint on profit seeking: Entitlements in the market.Amer . Econ. Rev.76, 728–741 (1986)

    1986

  34. [43]

    & Schwarze, B

    Güth, W., Schmittberger, R. & Schwarze, B. An experimental analysis of ultimatum bargaining.J. Econ. Behav. Organ.3, 367–388, DOI: 10.1016/0167-2681(82)90011-7 (1982)

    work page doi:10.1016/0167-2681(82)90011-7 1982

  35. [44]

    & McCabe, K

    Berg, J., Dickhaut, J. & McCabe, K. Trust, reciprocity, and social history.Games Econ. Behav.10, 122–142, DOI: 10.1006/game.1995.1027 (1995)

    arXiv 1995

  36. [45]

    & Page, S

    Hong, L. & Page, S. E. Groups of diverse problem solvers can outperform groups of high-ability problem solvers.Proc. Natl. Acad. Sci. USA101, 16385–16389, DOI: 10.1073/pnas.0403723101 (2004). 46.Milgram, S. Behavioral study of obedience.J. Abnorm. Soc. Psychol.67, 371–378, DOI: 10.1037/h0040525 (1963). 15/24 Acknowledgments ASC thanks the office of the vi...

    arXiv 2004

  37. [46]

    Since the monetary prize for each player is the number of points multiplied by 10, it is clear that earning more points results in a higher prize

  38. [47]

    - If you choose B: (B,B) gives you the lowest reward, and (B,A) gives a slightly higher reward, but still lower than all possible rewards when choosing A

    Comparing the outcomes in each group: - If you choose A: (A,A) gives you the highest reward, and (A,B) gives you the same reward but a lower-one for the other player. - If you choose B: (B,B) gives you the lowest reward, and (B,A) gives a slightly higher reward, but still lower than all possible rewards when choosing A

  39. [48]

    Even though one might worry about the other player choice, it doesn’t make sense since the only possible outcome of choosing A is 10 points, which is the maximum

    Based on this information, choosing A would result in a higher potential reward, making it the better choice. Even though one might worry about the other player choice, it doesn’t make sense since the only possible outcome of choosing A is 10 points, which is the maximum. The other player choice isn’t going to affect your outcome at all, there is no risk....