Pith. sign in

REVIEW 3 major objections 5 minor 71 references

Understanding Human Limits in Pattern Recognition: A Computational Model of Sequential Reasoning in Rock, Paper, Scissors

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 In rock-paper-scissors, the bottleneck is imagining the pattern.

desk verdict A solid LLM-based cognitive modeling paper that convincingly locates the bottleneck for the model in hypothesis generation, but the leap from model oracle results to human cognition is under-evidenced. read the letter →

arxiv 2508.06503 v1 pith:2FOF7MYD submitted 2025-07-25 q-bio.NC cs.AI

Logan Cross , Erik Brockbank , Tobias Gerstenberg , Judith E. Fan , Daniel L. K. Yamins , Nick Haber This is my paper
classification q-bio.NCcs.AI
keywords rockpaperscissorstheoryofmindlargelanguagemodelshypothesisgenerationpatternlearningsequentialreasoningcognitivemodelingopponent
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 argues that people fail to exploit complex sequential patterns in an opponent's play because they do not generate the right hypotheses about what the opponent is doing, not because they cannot evaluate or act on a good hypothesis. The authors reach this conclusion using Hypothetical Minds, an LLM-based agent that generates and scores natural-language hypotheses about opponent strategy, and show it reproduces human win rates against seven algorithmic opponents. When the model is given the true strategy in words, its win rate jumps above 80% for nearly every opponent; when it is given hints that direct attention to relevant contingencies, it also improves. The paper therefore identifies hypothesis generation as the primary cognitive bottleneck in sequential pattern recognition, and proposes that verbal scaffolding may help people overcome the same limitation.

What carries the argument

Hypothetical Minds (HM), an LLM-based agent that approximates Bayesian inference over opponent strategies. Its Theory of Mind module generates candidate hypotheses in natural language, scores each by recency-weighted prediction accuracy (a Rescorla-style value update with learning rate 0.3), validates hypotheses above a threshold, and feeds the best hypothesis to a chain-of-thought decision reflection that picks the counter-move. The load-bearing mechanism is the separation between generating hypotheses and evaluating them: augmentations that replace generation with oracle knowledge reveal that evaluation and planning are sufficient once the right hypothesis is available.

What would settle it

Give human participants the same natural-language strategy descriptions (or attention scaffolds) used in the model's Give Hypothesis and Attention Scaffold conditions. If participants still play at chance against the outcome-dependent bots—or if HM fails to exceed 80% win rate when given oracle descriptions of those same bots—the claim that hypothesis generation is the primary bottleneck would be falsified, pointing instead to evaluation or planning limits.

Watch

Extended reading notes

Core claim

The central claim is that in iterated rock-paper-scissors, human-like limits on exploiting structured opponents are driven by hypothesis generation, not hypothesis evaluation or move planning. The paper demonstrates this by first showing that Hypothetical Minds—which maintains a memory of past rounds, generates candidate strategies in natural language, scores them by prediction accuracy, and reflects before acting—matches human performance across six of seven bots, including the same selective failures. It then ablates and augments components: removing hypothesis evaluation or theory-of-mind collapses performance, while supplying the true strategy description (Give Hypothesis) or a menu of candidate strategies (Choose Hypothesis) lifts win rates above 80% for both transition-based and outcome-dependent bots. The one bot that resists even oracle knowledge is the most complex one, whose strategy needs nine condition-action pairs, suggesting a separate working-memory-style limit. The paper concludes that search over hypotheses—the space of strategies a player can imagine—is the main constraint, and that restructuring that space through attention-directing hints can release it.

Load-bearing premise

The argument transfers conclusions about the LLM agent's internal bottleneck to human cognition, relying on the assumption that GPT-4o's natural-language hypothesis distribution is a good stand-in for people's prior expectations about opponent strategies.

Editorial extensions

If this is right

  • If hypothesis generation is the bottleneck, then interventions that expand or redirect a player's candidate strategies—such as attention scaffolds or analogical examples—should improve exploitation of moderately complex patterns, exactly as the model shows.
  • Performance against the most complex opponent will not be rescued by merely giving the true strategy; the nine-condition rule exceeds the agent's (and plausibly humans') ability to implement it, pointing to a separate capacity limit.
  • Model-based comparison of LLM agents can serve as a testbed for cognitive theories: ablating components of HM produces failure patterns that distinguish hypothesis generation, evaluation, and planning accounts.
  • The specific failure of GPT-4o on the copying bot, and its success when the backbone is Llama 3, implies the bottleneck distribution is not a fixed property of the architecture but depends on the prior knowledge in the base model.

Reading between the lines

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

  • Editorial inference: If the same scaffolding manipulations were run with human participants and produced similar gains, it would support a practical pedagogy for improving pattern detection in adversarial settings; the paper's authors state this is a testable prediction, not something they demonstrated with humans.
  • Editorial inference: The dependency of the bottleneck on the base LLM's priors suggests that the 'hypothesis space' is not a stable cognitive universal; different populations or cultures may have different default strategy priors, which would change which patterns are easy or hard.
  • Editorial inference: A natural extension is to measure the hypothesis-generation bottleneck directly by eliciting free descriptions of opponent strategy from humans and comparing the distribution of proposed hypotheses to HM's, quantifying how much of the human failure is explained by missing hypotheses versus mis-evaluation.
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 uses Hypothetical Minds (HM), an LLM-based agent with a Theory-of-Mind module that generates and evaluates natural-language hypotheses about opponents, to model human performance in iterated rock-paper-scissors games from Brockbank & Vul (2024). Against seven algorithmic bots of increasing sequential complexity, HM is reported to reproduce the qualitative human pattern: success on simple transition-based bots and failure on outcome-dependent and combined-contingency bots. The authors then augment HM in Experiment 2 by providing oracle natural-language descriptions of the true opponent strategies ('Give Hypothesis') or a list of candidate strategies ('Choose Hypothesis'); the large performance gains in these conditions lead them to conclude that hypothesis generation, rather than hypothesis evaluation or move planning, is the primary cognitive bottleneck. Experiment 3 shows that increasing the number of maintained hypotheses or sampling temperature does not improve HM, whereas verbal attention scaffolds and analogical scaffolds improve performance on the outcome-dependent bots. The paper frames these results as suggesting that human pattern-recognition limits similarly stem from hypothesis generation, and proposes testable predictions for human experiments.

Significance. The manuscript's strengths are its systematic architecture comparison (HM vs. ablations and baselines), the oracle augmentation design, the exploration of base-LLM sensitivity, and the clear presentation of scaffolding manipulations. The code and data are available on GitHub, which supports reproducibility. The decomposition of model failure into hypothesis generation, evaluation, and planning is a useful framework, and the scaffolding results provide concrete, falsifiable predictions for human studies. However, the paper's central human-facing claim—that hypothesis generation is the primary cognitive bottleneck for people—is an extrapolation from LLM behavior that is not directly tested in humans. The quantitative fit between HM and human learning trajectories is modest (L2 = 2.95 vs. a human noise ceiling of 1.40), and the model's qualitative pattern depends on the base LLM, with GPT-4o failing on a bot that humans easily exploit. As a model-based analysis generating hypotheses about human cognition, the work is valuable; as a demonstration of human cognitive limits, it is under-supported.

major comments (3)
  1. [Experiment 2 and Discussion] The central claim that hypothesis generation, not evaluation or planning, is the primary cognitive bottleneck for humans is not directly supported by the evidence presented. The Give Hypothesis augmentation shows that HM, an LLM agent, can exploit explicit natural-language strategy descriptions, but no human experiment with strategy disclosure, scaffolding, or equivalent manipulation is reported. The Discussion explicitly concedes that 'this needs to be verified' and states 'we hypothesize that many humans would be able to perform well when given the hypothesis,' yet the Abstract and Experiment 2 text present the conclusion without this hedge. Because the paper's goal is 'Understanding Human Limits,' the human-facing interpretation should either be supported by a human experiment or consistently reframed as a model-derived prediction requiring validation.
  2. [Table 1, Figures 3-4] The quantitative correspondence between HM and human behavior is weak by the paper's own noise ceiling. Table 1 reports a trial-bin L2 distance of 2.95 between HM and humans, compared to a human-human split-half noise ceiling of 1.40, and a trajectory correlation of r=0.38 versus 0.81 for the human ceiling. This undercuts the description of HM as 'closely mirroring' human performance. Additionally, HM with GPT-4o performs below chance on the Opponent-transition(0) copying bot, which humans readily exploit (Figure 3); the paper shifts to Llama 3 for this opponent, but this demonstrates that the model's hypothesis distribution is base-LLM dependent rather than a stable cognitive universal. This dependence matters because the bottleneck conclusion is transferred from the model to humans via the assumed representativeness of its hypothesis priors.
  3. [Experiment 3] The claim that 'the challenge in hypothesis generation lies not in considering more possibilities, but in generating the right kinds of hypotheses in the first place' is tested only within HM. The top_h and temperature manipulations show that these specific interventions do not help the LLM agent, but there is no human analogue demonstrating that people also fail to benefit from wider search or additional samples. The paper's own language in Experiment 3 acknowledges this: 'This result introduces an intriguing hypothesis about cognition and learning that can then be tested in humans.' The bottleneck conclusion should be presented as a hypothesis about human cognition, not as an established finding, unless human data are provided.
minor comments (5)
  1. [Tables 1 and 2] The table captions are misnumbered: the table titled 'Table 1: Model-Human Similarity' is referred to in the text as 'Table 2' in several places, and the following table is also numbered 'Table 2.' The numbering should be corrected throughout.
  2. [Methods, bot notation] The shorthand names for the outcome-dependent bots (W0L+T− and W+L−T0) are introduced without an explicit legend; a brief explanation of the notation (e.g., 'W', 'L', 'T' followed by transition types) would help readers parse the figures and tables.
  3. [Experiment 1, Table 1] The claim that HM 'maintained significantly better alignment' with human learning patterns is not backed by any significance test or confidence interval for the L2 or correlation values; the word 'significantly' should be replaced with a quantitative comparison or a formal test across seeds.
  4. [Experiment 3] The assertion that 'performance is not sensitive to this value (Cross et al., 2024)' for V_thr is cited from prior work but not verified for the new RPS opponent set; a sensitivity check in this task would strengthen the claim.
  5. [Figure 4 caption] The caption 'LLM comparison' is uninformative; it should specify the performance measure (win rate), the error bars (SEM over three seeds), and the opponent types shown.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the bottleneck claim rests on new within-paper model manipulations, and the human transfer is explicitly hedged as a hypothesis to be tested.

full rationale

The derivation chain is self-contained. HM is applied to the seven RPS bot conditions with no parameter fitted to the human win rates from Brockbank & Vul (2024): hyperparameters (alpha=0.3, Vthr=0.7, top h=5, temperature) come from the prior HM paper or are fixed, and the model-human match is reported as an a posteriori similarity (Table 1, win-rate L2 = 0.53 versus a 0.17 human noise ceiling), not as a fitted prediction. The central claim that hypothesis generation, not evaluation or planning, is the bottleneck is established by model-only manipulations whose logic does not reduce to the input data: removing hypothesis evaluation degrades performance, and replacing the ToM module with oracle strategy text (Give Hypothesis) or with the fixed set of seven candidate strategies (Choose Hypothesis) raises win rates above 80% on 6/7 bots. Because the oracle text is the ground-truth strategy and the manipulation is causally upstream of the measured win rate, this is a genuine experimental dissociation rather than a definitional equivalence. The human-facing conclusion is an analogical transfer that the paper itself flags as unverified: "However, this needs to be verified and we hypothesize that many humans would be able to perform well when given the hypothesis" (Discussion) and "This result introduces an intriguing hypothesis about cognition and learning that can then be tested in humans" (Experiment 3). That is an open empirical question and a validity risk, not a circular derivation. Self-citations are present (HM from Cross et al. 2024; human data from Brockbank & Vul 2024, both with overlapping authorship), but the load-bearing bottleneck claim rests on this paper's own ablations, augmentations, scaffolding, and temperature manipulations rather than on those citations. No step in the paper exhibits a reduction of an output to its input by construction, so the circularity score is low.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claims rest on modeling choices imported from prior work by the same authors (Cross et al., 2024; Brockbank & Vul, 2024). No parameter is fitted to the human RPS data, but the LLM backbone is selected by closeness to human behavior, and the scaffold prompts are hand-authored. The most consequential unvalidated assumption is that GPT-4o's natural-language hypothesis prior approximates human priors.

free parameters (4)
  • learning rate alpha = 0.3
    Controls the Rescorla-Wagner update of hypothesis values; imported from Cross et al. (2024), no sensitivity analysis in this paper.
  • validation threshold V_thr = 0.7
    Hypothesis is validated when its value exceeds 0.7; authors claim insensitivity from prior work but do not show it here.
  • top_h (number of maintained hypotheses) = 5
    Default pool of top hypotheses; tested range 0 to 9, performance is flat above 1, so this is not a critical parameter.
  • LLM sampling temperature = 0.2
    Default temperature; increasing to 1.0 degrades performance, showing sensitivity to this choice.
assumptions (5)
  • domain assumption Opponent strategies are expressible as natural-language hypotheses that an LLM can generate and evaluate.
    The ToM module prompts the LLM to produce hypotheses in language; the paper argues language is a useful domain-general parameterization (Introduction, Methods).
  • domain assumption Win rate against a bot reflects the quality of a player's predictive model of the opponent.
    The paper uses win rate as the primary performance metric throughout; this assumes better prediction translates directly to better moves given the payoff structure.
  • domain assumption The bots' 90% dependency probability and payoff structure are taken as ground truth from Brockbank & Vul (2024).
    Human data and bot definitions come from the cited prior study; any errors there propagate.
  • ad hoc to paper HM's hyperparameters (alpha=0.3, V_thr=0.7) from Cross et al. (2024) are appropriate for RPS without re-estimation.
    The authors import these values from earlier work; they note V_thr is insensitive but do not test alpha here.
  • ad hoc to paper The natural-language descriptions of the true strategies in the Give Hypothesis condition faithfully capture the bots' rules.
    These descriptions were written by the authors; the most complex description is 352 words and may itself be a source of difficulty.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Understanding Human Limits in Pattern Recognition: A Computational Model of Sequential Reasoning in Rock, Paper, Scissors." pith.science (2026). https://pith.science/paper/2FOF7MYD

@misc{pith2026250806503,
  author       = {Pith},
  title        = {Pith review of: Understanding Human Limits in Pattern Recognition: A Computational Model of Sequential Reasoning in Rock, Paper, Scissors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2FOF7MYD}},
  note         = {Machine review of arXiv:2508.06503}
}
read the original abstract

How do we predict others from patterns in their behavior and what are the computational constraints that limit this ability? We investigate these questions by modeling human behavior over repeated games of rock, paper, scissors from Brockbank & Vul (2024). Against algorithmic opponents that varied in strategic sophistication, people readily exploit simple transition patterns (e.g., consistently playing rock after paper) but struggle to detect more complex sequential dependencies. To understand the cognitive mechanisms underlying these abilities and their limitations, we deploy Hypothetical Minds (HM), a large language model-based agent that generates and tests hypotheses about opponent strategies, as a cognitive model of this behavior (Cross et al., 2024). We show that when applied to the same experimental conditions, HM closely mirrors human performance patterns, succeeding and failing in similar ways. To better understand the source of HM's failures and whether people might face similar cognitive bottlenecks in this context, we performed a series of ablations and augmentations targeting different components of the system. When provided with natural language descriptions of the opponents' strategies, HM successfully exploited 6/7 bot opponents with win rates >80% suggesting that accurate hypothesis generation is the primary cognitive bottleneck in this task. Further, by systematically manipulating the model's hypotheses through pedagogically-inspired interventions, we find that the model substantially updates its causal understanding of opponent behavior, revealing how model-based analyses can produce testable hypotheses about human cognition.

Figures

Figures reproduced from arXiv: 2508.06503 by the authors.

Figure 1
Figure 1. Architecture of Hypothetical Minds and Sequential algorithmic opponents. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Human performance against sequential opponents. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. HM baseline model performance. Error bars reflect SEM over 3 seeds. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: LLM comparison tlenecks beyond hypothesis generation, namely, the cognitive demands of accurately reasoning about and implementing re￾sponses to multiple nested contingencies. Experiment 3: Improving hypothesis generation Increasing the number of hypotheses evaluated d…
Figure 5
Figure 5. Figure 5: Augmentations for hypothesis generation and hy [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Varied number of hypotheses and LLM temperature [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Result of scaffolding interventions on hypothesis [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: LLM architectures in model comparison [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: LLM architectures in model augmentations. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 53 canonical work pages

  1. [1]

    \ Schrater, P R

    acuna2008structure APACrefauthors Acuna, D. \ Schrater, P R. APACrefauthors \ 2008 . Structure learning in human sequential decision-making Structure learning in human sequential decision-making . Advances in neural information processing systems 21

    work page 2008

  2. [2]

    , Br \"a ndle, F

    allen2024using APACrefauthors Allen, K. , Br \"a ndle, F. , Botvinick, M. , Fan, J E. , Gershman, S J. , Gopnik, A. others APACrefauthors \ 2024 . Using games to understand the mind Using games to understand the mind . Nature Human Behaviour 1--9

    work page 2024

  3. [3]

    APACrefauthors \ 1991

    anderson1991adaptive APACrefauthors Anderson, J R. APACrefauthors \ 1991 . The adaptive nature of human categorization. The adaptive nature of human categorization. Psychological review 98 3 409

    work page 1991

  4. [4]

    , Jara-Ettinger, J

    baker2017rational APACrefauthors Baker, C L. , Jara-Ettinger, J. , Saxe, R. \ Tenenbaum, J B. APACrefauthors \ 2017 . Rational quantitative attribution of beliefs, desires and percepts in human mentalizing Rational quantitative attribution of beliefs, desires and percepts in human mentalizing . Nature Human Behaviour 1 4 0064

    work page 2017

  5. [5]

    , Saxe, R

    baker2009action APACrefauthors Baker, C L. , Saxe, R. \ Tenenbaum, J B. APACrefauthors \ 2009 . Action understanding as inverse planning Action understanding as inverse planning . Cognition 113 3 329--349

    work page 2009

  6. [6]

    , Jaffe, S

    batzilis2019behavior APACrefauthors Batzilis, D. , Jaffe, S. , Levitt, S. , List, J A. \ Picel, J. APACrefauthors \ 2019 . Behavior in strategic settings: Evidence from a million rock-paper-scissors games Behavior in strategic settings: Evidence from a million rock-paper-scissors games . Games 10 2 18

    work page 2019

  7. [7]

    APACrefauthors \ 2014

    belland2014scaffolding APACrefauthors Belland, B R. APACrefauthors \ 2014 . Scaffolding: Definition, current debates, and future directions Scaffolding: Definition, current debates, and future directions . Handbook of research on educational communications and technology 505--518

    work page 2014

  8. [8]

    , Dasgupta, I

    binz2024meta APACrefauthors Binz, M. , Dasgupta, I. , Jagadish, A K. , Botvinick, M. , Wang, J X. \ Schulz, E. APACrefauthors \ 2024 . Meta-learned models of cognition Meta-learned models of cognition . Behavioral and Brain Sciences 47 e147

    work page 2024

Show all 71 references
  1. [9]

    \ Schulz, E

    binz2023turning APACrefauthors Binz, M. \ Schulz, E. APACrefauthors \ 2023 . Turning large language models into cognitive models Turning large language models into cognitive models . arXiv preprint arXiv:2306.03917

    2023 arXiv

  2. [10]

    , Dayan, P

    bramley2017formalizing APACrefauthors Bramley, N R. , Dayan, P. , Griffiths, T L. \ Lagnado, D A. APACrefauthors \ 2017 . Formalizing Neurath's ship: Approximate algorithms for online causal learning. Formalizing neurath's ship: Approximate algorithms for online causal learnin...

    2017

  3. [11]

    bramley2023active APACrefauthors Bramley, N R. \ Xu, F. APACrefauthors \ 2023 . Active inductive inference in children and adults: A constructivist perspective Active inductive inference in children and adults: A constructivist perspective . Cognition 238 105471

    2023

  4. [12]

    \ Vul, E

    brockbank2021formalizing APACrefauthors Brockbank, E. \ Vul, E. APACrefauthors \ 2021 . Formalizing opponent modeling with the rock, paper, scissors game Formalizing opponent modeling with the rock, paper, scissors game . Games 12 3 70

    2021

  5. [13]

    \ Vul, E

    brockbank2024repeated APACrefauthors Brockbank, E. \ Vul, E. APACrefauthors \ 2024 . Repeated rock, paper, scissors play reveals limits in adaptive sequential behavior Repeated rock, paper, scissors play reveals limits in adaptive sequential behavior . Cognitive Psychology 151 101654

    2024

  6. [14]

    APACrefauthors \ 2011

    camerer2011behavioral APACrefauthors Camerer, C F. APACrefauthors \ 2011 . Behavioral game theory: Experiments in strategic interaction Behavioral game theory: Experiments in strategic interaction . Princeton university press

    2011

  7. [15]

    , Xiang, V

    cross2024hypothetical APACrefauthors Cross, L. , Xiang, V. , Bhatia, A. , Yamins, D L. \ Haber, N. APACrefauthors \ 2024 . Hypothetical minds: Scaffolding theory of mind for multi-agent tasks with large language models Hypothetical minds: Scaffolding theory of mind for multi-a...

    2024 arXiv

  8. [16]

    \ Tobler, P N

    daw2014value APACrefauthors Daw, N D. \ Tobler, P N. APACrefauthors \ 2014 . Value learning through reinforcement: the basics of dopamine and reinforcement learning Value learning through reinforcement: the basics of dopamine and reinforcement learning . Neuroeconomics Neuroec...

    2014

  9. [17]

    APACrefauthors \ 2019

    dyson2019 APACrefauthors Dyson, B J. APACrefauthors \ 2019 . Behavioural isomorphism, cognitive economy and recursive thought in non-transitive game strategy Behavioural isomorphism, cognitive economy and recursive thought in non-transitive game strategy . Games 10 3 32

    2019

  10. [18]

    \ Nassar, M R

    feldmanhall2021computational APACrefauthors FeldmanHall, O. \ Nassar, M R. APACrefauthors \ 2021 . The computational challenge of social learning The computational challenge of social learning . Trends in Cognitive Sciences 25 12 1045--1057

    2021

  11. [19]

    \ Shenhav, A

    feldmanhall2019resolving APACrefauthors FeldmanHall, O. \ Shenhav, A. APACrefauthors \ 2019 . Resolving uncertainty in a social world Resolving uncertainty in a social world . Nature human behaviour 3 5 426--435

    2019

  12. [20]

    \ Dyson, B J

    forder_dyson2016 APACrefauthors Forder, L. \ Dyson, B J. APACrefauthors \ 2016 . Behavioural and neural modulation of win-stay but not lose-shift strategies as a function of outcome value in Rock, Paper, Scissors Behavioural and neural modulation of win-stay but not lose-shift...

    2016

  13. [21]

    , Theodoropoulos, N C

    franken2022algorithms APACrefauthors Fr \"a nken, J P. , Theodoropoulos, N C. \ Bramley, N R. APACrefauthors \ 2022 . Algorithms of adaptation in inductive inference Algorithms of adaptation in inductive inference . Cognitive Psychology 137 101506

    2022

  14. [22]

    \ Hoyos, C

    gentner2017analogy APACrefauthors Gentner, D. \ Hoyos, C. APACrefauthors \ 2017 . Analogy and abstraction Analogy and abstraction . Topics in cognitive science 9 3 672--693

    2017

  15. [23]

    \ Niv, Y

    gershman2013perceptual APACrefauthors Gershman, S J. \ Niv, Y. APACrefauthors \ 2013 . Perceptual estimation obeys Occam's razor Perceptual estimation obeys occam's razor . Frontiers in psychology 4 623

    2013

  16. [24]

    , Norman, K A

    gershman2015discovering APACrefauthors Gershman, S J. , Norman, K A. \ Niv, Y. APACrefauthors \ 2015 . Discovering latent causes in reinforcement learning Discovering latent causes in reinforcement learning . Current Opinion in Behavioral Sciences 5 43--50

    2015

  17. [25]

    \ Speekenbrink, M

    guennouni2022transfer APACrefauthors Guennouni, I. \ Speekenbrink, M. APACrefauthors \ 2022 . Transfer of learned opponent models in zero sum games Transfer of learned opponent models in zero sum games . Computational Brain & Behavior 5 3 326--342

    2022

  18. [26]

    , Yang, D

    guo2025deepseek APACrefauthors Guo, D. , Yang, D. , Zhang, H. , Song, J. , Zhang, R. , Xu, R. others APACrefauthors \ 2025 . Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning Deepseek-r1: Incentivizing reasoning capability in llms via reinforce...

    2025 arXiv

  19. [27]

    , Saxe, R

    ho2022planning APACrefauthors Ho, M K. , Saxe, R. \ Cushman, F. APACrefauthors \ 2022 . Planning with theory of mind Planning with theory of mind . Trends in Cognitive Sciences 26 11 959--971

    2022

  20. [28]

    , Karuza, E A

    kahn2018network APACrefauthors Kahn, A E. , Karuza, E A. , Vettel, J M. \ Bassett, D S. APACrefauthors \ 2018 . Network constraints on learnability of probabilistic motor sequences Network constraints on learnability of probabilistic motor sequences . Nature human behaviour 2 ...

    2018

  21. [29]

    , Jeong, J

    kang2025t1 APACrefauthors Kang, M. , Jeong, J. \ Cho, J. APACrefauthors \ 2025 . T1: Tool-integrated Self-verification for Test-time Compute Scaling in Small Language Models T1: Tool-integrated self-verification for test-time compute scaling in small language models . arXiv pr...

    2025 arXiv

  22. [30]

    , Ho, M K

    kleiman2016coordinate APACrefauthors Kleiman-Weiner, M. , Ho, M K. , Austerweil, J L. , Littman, M L. \ Tenenbaum, J B. APACrefauthors \ 2016 . Coordinate to cooperate or compete: abstract goals and joint intentions in social interaction Coordinate to cooperate or compete: abs...

    2016

  23. [31]

    , Dong, G

    li2025search APACrefauthors Li, X. , Dong, G. , Jin, J. , Zhang, Y. , Zhou, Y. , Zhu, Y. Dou, Z. APACrefauthors \ 2025 . Search-o1: Agentic search-enhanced large reasoning models Search-o1: Agentic search-enhanced large reasoning models . arXiv preprint arXiv:2501.05366

    2025 arXiv

  24. [32]

    \ Griffiths, T L

    lieder2020resource APACrefauthors Lieder, F. \ Griffiths, T L. APACrefauthors \ 2020 . Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources Resource-rational analysis: Understanding human cognition as the optimal use o...

    2020

  25. [33]

    \ Singh, C

    lin2011using APACrefauthors Lin, S Y. \ Singh, C. APACrefauthors \ 2011 . Using isomorphic problems to learn introductory physics Using isomorphic problems to learn introductory physics . Physical Review Special Topics—Physics Education Research 7 2 020104

    2011

  26. [34]

    , Sinha, D A

    loya2023exploring APACrefauthors Loya, M. , Sinha, D A. \ Futrell, R. APACrefauthors \ 2023 . Exploring the Sensitivity of LLMs' Decision-Making Capabilities: Insights from Prompt Variation and Hyperparameters Exploring the sensitivity of llms' decision-making capabilities: In...

    2023 arXiv

  27. [35]

    \ Neumann, J V

    von_neumann APACrefauthors Morgenstern, O. \ Neumann, J V. APACrefauthors \ 1953 . Theory of games and economic behavior Theory of games and economic behavior . Princeton university press

    1953

  28. [36]

    , Yang, Z

    muennighoff2025s1 APACrefauthors Muennighoff, N. , Yang, Z. , Shi, W. , Li, X L. , Fei-Fei, L. , Hajishirzi, H. Hashimoto, T. APACrefauthors \ 2025 . s1: Simple test-time scaling s1: Simple test-time scaling . arXiv preprint arXiv:2501.19393

    2025 arXiv

  29. [37]

    , Hilton, J

    nakano2021webgpt APACrefauthors Nakano, R. , Hilton, J. , Balaji, S. , Wu, J. , Ouyang, L. , Kim, C. others APACrefauthors \ 2021 . Webgpt: Browser-assisted question-answering with human feedback Webgpt: Browser-assisted question-answering with human feedback . arXiv preprint ...

    2021 arXiv

  30. [38]

    APACrefauthors \ 1950

    nash APACrefauthors Nash, J F. APACrefauthors \ 1950 . Equilibrium points in n-person games Equilibrium points in n-person games . Proceedings of the national academy of sciences 36 1 48--49

    1950

  31. [39]

    , Zaki, J

    ong2019computational APACrefauthors Ong, D C. , Zaki, J. \ Goodman, N D. APACrefauthors \ 2019 . Computational models of emotion inference in theory of mind: A review and roadmap Computational models of emotion inference in theory of mind: A review and roadmap . Topics in cogn...

    2019

  32. [40]

    , Zou, C Q

    park2024generative APACrefauthors Park, J S. , Zou, C Q. , Shaw, A. , Hill, B M. , Cai, C. , Morris, M R. Bernstein, M S. APACrefauthors \ 2024 . Generative agent simulations of 1,000 people Generative agent simulations of 1,000 people . arXiv preprint arXiv:2411.10109

    2024 arXiv

  33. [41]

    , Broman, D

    poesia2024learning APACrefauthors Poesia, G. , Broman, D. , Haber, N. \ Goodman, N. APACrefauthors \ 2024 . Learning formal mathematics from intrinsic motivation Learning formal mathematics from intrinsic motivation . Advances in Neural Information Processing Systems 37 43032--43057

    2024

  34. [42]

    \ Budescu, D V

    rapoport1997 APACrefauthors Rapoport, A. \ Budescu, D V. APACrefauthors \ 1997 . Randomization in individual choice behavior Randomization in individual choice behavior . Psychological Review 104 3 603--617

    1997

  35. [43]

    APACrefauthors \ 1972

    rescorla1972theory APACrefauthors Rescorla, R A. APACrefauthors \ 1972 . A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and non-reinforcement A theory of pavlovian conditioning: Variations in the effectiveness of reinforcement and non-rein...

    1972

  36. [44]

    , Steixner-Kumar, S

    rusch2020theory APACrefauthors Rusch, T. , Steixner-Kumar, S. , Doshi, P. , Spezio, M. \ Gl \"a scher, J. APACrefauthors \ 2020 . Theory of mind and decision science: Towards a typology of tasks and computational models Theory of mind and decision science: Towards a typology o...

    2020

  37. [45]

    , Griffiths, T L

    sanborn2010rational APACrefauthors Sanborn, A N. , Griffiths, T L. \ Navarro, D J. APACrefauthors \ 2010 . Rational approximations to rational models: alternative algorithms for category learning. Rational approximations to rational models: alternative algorithms for category ...

    2010

  38. [46]

    , Choi, Y

    sclar2023quantifying APACrefauthors Sclar, M. , Choi, Y. , Tsvetkov, Y. \ Suhr, A. APACrefauthors \ 2023 . Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting Quantifying language models' se...

    2023 arXiv

  39. [47]

    , St \"o ttinger, E

    danckert_2012_modeling APACrefauthors Sepahvand, N M. , St \"o ttinger, E. , Danckert, J. \ Anderson, B. APACrefauthors \ 2014 . Sequential decisions: a computational comparison of observational and reinforcement accounts Sequential decisions: a computational comparison of obs...

    2014

  40. [48]

    , Friedman, D

    si2023measuring APACrefauthors Si, C. , Friedman, D. , Joshi, N. , Feng, S. , Chen, D. \ He, H. APACrefauthors \ 2023 . Measuring inductive biases of in-context learning with underspecified demonstrations Measuring inductive biases of in-context learning with underspecified de...

    2023 arXiv

  41. [49]

    APACrefauthors \ 1972

    simon1972theories APACrefauthors Simon, H A. APACrefauthors \ 1972 . Theories of bounded rationality Theories of bounded rationality . Decision and organization 1 1 161--176

    1972

  42. [50]

    APACrefauthors \ 2008

    singh2008assessing APACrefauthors Singh, C. APACrefauthors \ 2008 . Assessing student expertise in introductory physics with isomorphic problems. II. Effect of some potential factors on problem solving and transfer Assessing student expertise in introductory physics with isomo...

    2008

  43. [51]

    , Filipowicz, A

    prior_learned_strategies APACrefauthors St \"o ttinger, E. , Filipowicz, A. , Danckert, J. \ Anderson, B. APACrefauthors \ 2014 . The effects of prior learned strategies on updating an opponent's strategy in the rock, paper, scissors game The effects of prior learned strategie...

    2014

  44. [52]

    \ Thornton, M A

    tamir2018modeling APACrefauthors Tamir, D I. \ Thornton, M A. APACrefauthors \ 2018 . Modeling the predictive social mind Modeling the predictive social mind . Trends in cognitive sciences 22 3 201--212

    2018

  45. [53]

    \ Tamir, D I

    thornton2017mental APACrefauthors Thornton, M A. \ Tamir, D I. APACrefauthors \ 2017 . Mental models accurately predict emotion transitions Mental models accurately predict emotion transitions . Proceedings of the National Academy of Sciences 114 23 5982--5987

    2017

  46. [54]

    \ Tamir, D I

    thornton2021people APACrefauthors Thornton, M A. \ Tamir, D I. APACrefauthors \ 2021 . People accurately predict the transition probabilities between actions People accurately predict the transition probabilities between actions . Science Advances 7 9 eabd4995

    2021

  47. [55]

    , Baker, C

    ullman2009help APACrefauthors Ullman, T. , Baker, C. , Macindoe, O. , Evans, O. , Goodman, N. \ Tenenbaum, J. APACrefauthors \ 2009 . Help or hinder: Bayesian models of social goal inference Help or hinder: Bayesian models of social goal inference . Advances in neural informat...

    2009

  48. [56]

    , Nassar, M R

    van2022latent APACrefauthors van Baar, J M. , Nassar, M R. , Deng, W. \ FeldmanHall, O. APACrefauthors \ 2022 . Latent motives guide structure learning during adaptive social choice Latent motives guide structure learning during adaptive social choice . Nature Human Behaviour ...

    2022

  49. [57]

    , Agapiou, J P

    vezhnevets2023generative APACrefauthors Vezhnevets, A S. , Agapiou, J P. , Aharon, A. , Ziv, R. , Matyas, J. , Du \'e \ n ez-Guzm \'a n, E A. Leibo, J Z. APACrefauthors \ 2023 . Generative agent-based modeling with actions grounded in physical, social, or digital space using C...

    2023 arXiv

  50. [58]

    APACrefauthors \ 1978

    vygotsky1978mind APACrefauthors Vygotsky, L S. APACrefauthors \ 1978 . Mind in society: The development of higher psychological processes Mind in society: The development of higher psychological processes \ ( 86). Harvard university press

    1978

  51. [59]

    wang2023shepherd APACrefauthors Wang, T. , Yu, P. , Tan, X E. , O'Brien, S. , Pasunuru, R. , Dwivedi-Yu, J. Celikyilmaz, A. APACrefauthors \ 2023 . Shepherd: A critic for language model generation Shepherd: A critic for language model generation . arXiv preprint arXiv:2308.04592

    2023 arXiv

  52. [60]

    \ Pavlick, E

    webson2021prompt APACrefauthors Webson, A. \ Pavlick, E. APACrefauthors \ 2021 . Do prompt-based models really understand the meaning of their prompts? Do prompt-based models really understand the meaning of their prompts? arXiv preprint arXiv:2109.01247

    2021 arXiv

  53. [61]

    , Wang, X

    wei2022chain APACrefauthors Wei, J. , Wang, X. , Schuurmans, D. , Bosma, M. , Ichter, B. , Xia, F. Zhou, D. APACrefauthors \ 2022 . Chain-of-thought prompting elicits reasoning in large language models Chain-of-thought prompting elicits reasoning in large language models . Adv...

    2022

  54. [62]

    \ Lebiere, C

    west2001simple APACrefauthors West, R L. \ Lebiere, C. APACrefauthors \ 2001 . Simple games as dynamic, coupled systems: Randomness and other emergent properties Simple games as dynamic, coupled systems: Randomness and other emergent properties . Cognitive Systems Research 1 4...

    2001

  55. [63]

    , Grand, G

    wong2023word APACrefauthors Wong, L. , Grand, G. , Lew, A K. , Goodman, N D. , Mansinghka, V K. , Andreas, J. \ Tenenbaum, J B. APACrefauthors \ 2023 . From word models to world models: Translating from natural language to the probabilistic language of thought From word models...

    2023 arXiv

  56. [64]

    , Bruner, J S

    wood1976role APACrefauthors Wood, D. , Bruner, J S. \ Ross, G. APACrefauthors \ 1976 . The role of tutoring in problem solving The role of tutoring in problem solving . Journal of child psychology and psychiatry 17 2 89--100

    1976

  57. [65]

    , Sridhar, S

    wu2023computational APACrefauthors Wu, S A. , Sridhar, S. \ Gerstenberg, T. APACrefauthors \ 2023 . A computational model of responsibility judgments from counterfactual simulations and intention inferences A computational model of responsibility judgments from counterfactual ...

    2023

  58. [66]

    , Snell, C

    xiang2025towards APACrefauthors Xiang, V. , Snell, C. , Gandhi, K. , Albalak, A. , Singh, A. , Blagden, C. others APACrefauthors \ 2025 . Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Though Towards system 2 reasoning in llms: Learning how to thi...

    2025 arXiv

  59. [67]

    , V \'e lez, N

    xiang2023collaborative APACrefauthors Xiang, Y. , V \'e lez, N. \ Gershman, S J. APACrefauthors \ 2023 . Collaborative decision making is grounded in representations of other people’s competence and effort. Collaborative decision making is grounded in representations of other ...

    2023

  60. [68]

    yao2023tree APACrefauthors Yao, S. , Yu, D. , Zhao, J. , Shafran, I. , Griffiths, T. , Cao, Y. \ Narasimhan, K. APACrefauthors \ 2023 . Tree of thoughts: Deliberate problem solving with large language models Tree of thoughts: Deliberate problem solving with large language mode...

    2023

  61. [69]

    , Zhao, J

    yao2022react APACrefauthors Yao, S. , Zhao, J. , Yu, D. , Du, N. , Shafran, I. , Narasimhan, K. \ Cao, Y. APACrefauthors \ 2022 . React: Synergizing reasoning and acting in language models React: Synergizing reasoning and acting in language models . arXiv preprint arXiv:2210.03629

    2022 arXiv

  62. [70]

    , Moisan, F

    zhang2021rock APACrefauthors Zhang, H. , Moisan, F. \ Gonzalez, C. APACrefauthors \ 2021 . Rock-paper-scissors play: Beyond the win-stay/lose-change strategy Rock-paper-scissors play: Beyond the win-stay/lose-change strategy . Games 12 3 52

    2021

  63. [71]

    , Lucas, C G

    zhao2024model APACrefauthors Zhao, B. , Lucas, C G. \ Bramley, N R. APACrefauthors \ 2024 . A model of conceptual bootstrapping in human cognition A model of conceptual bootstrapping in human cognition . Nature Human Behaviour 8 1 125--136

    2024

Pith tools

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