Pith. sign in

REVIEW 3 major objections 4 minor 31 references

A reflective language-agent loop guided by human acceptance and calibration models raises successful, constraint-satisfying recommendations to 58–75 percent in a few iterations.

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 · grok-4.5

2026-07-12 05:22 UTC pith:BZTW2EDR

load-bearing objection Clean architecture that folds human-behavior models into a Reflexion loop with a proved termination bound; the numbers are real but rest on simulated humans and an LLM evaluator that both steers and scores. the 3 major comments →

arxiv 2607.03025 v1 pith:BZTW2EDR submitted 2026-07-03 cs.AI cs.MA

Human-Centric Reflective Architecture for Human-AI Collaborative Decision-Making

Andreas Kouridakis , Dimitrios Patiniotis Spyropoulos , George Vouros This is my paper
classification cs.AI cs.MA
keywords Agentic AIHuman-centric AICollaborative Decision MakingLarge Language ModelsLanguage AgentsReinforcement LearningAlignmentReflective Architecture
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.

Humans often over- or under-trust LLM advice, and raw accuracy alone does not produce recommendations people will actually accept. This paper treats the collaboration as a stochastic game whose objective is to minimize expected human loss, then builds an architecture that keeps reflecting—via linguistic feedback—until a termination condition derived from that loss is met. The architecture inserts two learned human models into the loop: one that transforms the agent’s raw confidence into a human-calibrated score, and one that predicts the probability a person will accept the current recommendation. Experiments in a tourism domain show that the full system reaches correct, constraint-respecting terminations far more often than a pure reflection baseline or ablations that drop the human models, typically after only four or five rounds, and that long-term memory further lifts success on new, more complex queries.

Core claim

Integrating human-calibrated confidence and acceptance models into a Reflexion-style linguistic loop, whose termination is governed by a derived human-loss bound, produces substantially higher rates of successful, constraint-satisfying recommendations while keeping the average number of iterations near four to five.

What carries the argument

Human-Centric Reflective Architecture (HCRA): a closed loop of actor, evaluator, self-reflection, short- and long-term memory, plus a human calibration model and a human acceptance model, whose joint objective is the human-loss termination condition of Theorem 3.1.

Load-bearing premise

That a small neural net trained on a re-balanced public multi-task dataset, together with an LLM evaluator that never sees ground-truth correctness, is a faithful enough stand-in for real human acceptance and factual accuracy that the measured success rates will hold with live users.

What would settle it

Replace the simulated human models with real users who see the calibrated confidences and accept or reject recommendations online; if the success rate then falls to the level of the uncalibrated baseline or the number of iterations balloons, the central claim fails.

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

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

3 major / 4 minor

Summary. The paper models human-AI collaborative decision-making as a stochastic game between an AI agent and a human (represented by behavior models), defines a human-utility objective via expected loss L_T, and derives a termination condition (Eq. 1) with a supporting theorem (Theorem 3.1). It proposes HCRA, which embeds human acceptance and calibration models into a Reflexion-style loop of actor, evaluator, and self-reflection LLMs that exchange linguistic feedback and use short/long-term memory. Experiments on 32+10 tourism recommendation questions (DeepSeek-V3) report 58.4% successful terminations (rising to 75% with long-term memory) at ~4.8 iterations, with ablations isolating the calibration model and evaluator reliability; a baseline without human models terminates faster but ignores constraints and acceptance.

Significance. If the claims hold, HCRA offers a concrete, test-time architecture that couples human-calibrated confidence transformation with linguistic reflection to raise constraint-satisfying acceptance rates while providing formal termination guarantees under the stated loss. Strengths include the explicit game formulation, the proof of Theorem 3.1 (Appendix B), open code/datasets, systematic temperature and ablation studies (Tables 1–4, Figures 5–8), and the demonstration that long-term memory improves performance on held-out complex queries. These elements advance human-centric agentic AI beyond pure RLHF or single-shot confidence calibration.

major comments (3)
  1. Section 6 defines a successful termination as acceptance probability >0.5 plus correctness and agreement; both šCorr_t and šAggr_t are produced by the same LLM evaluator that steers the self-reflection loop and feeds the human-acceptance model. No external ground-truth oracle is used despite the tourism domain admitting objective checks (venue existence, distances, prices, accessibility). Table 4 shows that injecting noise into the evaluator collapses success to 15.9%, confirming that headline rates (58–75%) are tightly coupled to evaluator reliability and therefore self-referential. An independent factual verifier (or human annotation of a subset) is required before the transfer claim can be accepted.
  2. The human acceptance and calibration models (Section 5, Appendix A) are trained on a re-balanced subset of the multi-task Vodrahalli et al. (2022) dataset and then frozen. Success is scored by the same acceptance model that participates in the loop. While the models themselves are external, the evaluation remains circular with respect to the simulated human; no live-user study or domain-matched human data for tourism is reported. This undermines the claim that the architecture “enhances decision-making effectiveness” for real users (Abstract, Section 7).
  3. Theorem 3.1 and Eq. (1) guarantee termination for any ε≥0 under the defined loss, yet the paper itself notes that termination need not be successful (erroneous but constraint-satisfying recommendations can be accepted). The experimental success metric therefore rests on an additional, unproven assumption that the LLM evaluator’s binary assessments correlate with true correctness. Without quantifying that correlation on the tourism questions, the formal guarantee does not underwrite the reported quality gains.
minor comments (4)
  1. Figure 1 numbering of component execution order is helpful but the prompt templates (Figure 2 / Appendix D) are only partially reproduced; full reproducible prompts should be included or linked.
  2. Notation for assessed quantities (šCorr_t, šAggr_t) is inconsistent across text and figures; a single glossary would improve readability.
  3. Appendix A.1 balancing procedure (75/25 ratio) is described but the exact sampling seed or code path is not stated; reproducibility would benefit from an explicit script reference.
  4. Table 1 reports average loss with standard deviations, yet the main text never discusses whether the observed differences across temperatures are statistically significant.

Circularity Check

1 steps flagged

Mild self-referential scoring: success is defined using the same LLM evaluator that steers the loop, but human models and the loss bound are independently derived from external data and definitions.

specific steps
  1. self definitional [Section 6 (success definition) + Evaluator (Section 4) + Ablation Table 4]
    "A termination is considered successful if at the final time step the acceptance probability is greater than 0.5, the recommendation is correct and in agreement to the constraints. ... The evaluator is an LLM that assesses the correctness and the agreement of the recommendation Re_t. ... The significant drop in success rate highlights the crucial role of the evaluator in HCRA: An unreliable evaluator adds noise in the reflective process, failing to steer HCRA to correct recommendations. [Table 4: Success Rate 15.9 %]"

    Success is defined by the evaluator's own binary assessments of correctness and agreement. Those same assessments are fed back into the self-reflection model and the human-acceptance model that steer the loop. The ablation that corrupts the evaluator collapses success to 15.9 %, confirming that the headline metric is produced by the same component that drives the process. No external ground-truth oracle is used despite the domain admitting one; the reported rates are therefore partly self-referential.

full rationale

The paper's central derivation (stochastic-game formulation, human-loss L_T, termination condition Eq. 1, Theorem 3.1) is self-contained: the loss is defined from the acceptance probability and the assessed class c(s_t), and the termination inequality follows algebraically without fitting to the tourism results. Human acceptance and calibration models are trained once on a re-balanced external dataset (Vodrahalli et al. 2022) and then frozen; they are not re-fit on the evaluation questions. The only mild circularity is that 'successful termination' (Section 6) is scored by the same LLM evaluator that supplies šCorr_t and šAggr_t to the self-reflection loop and to the human-acceptance model. Because the tourism domain admits objective ground truth (venue existence, distances, prices, accessibility) yet none is used, the reported 58–75 % success rates are partly self-referential. This is a scoring circularity, not a definitional or fitted-input circularity of the claimed derivation; the architecture and the loss bound remain independent of the evaluation metric. Hence score 2.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 2 invented entities

The central empirical claim rests on a small set of free parameters (epsilon, temperatures, alpha/beta of the calibration sigmoid, network widths) chosen by hand or early-stopped on the acceptance dataset, plus standard modeling assumptions that an LLM evaluator and a frozen human-behavior network adequately proxy real users and factual correctness. No new physical entities are postulated; the invented constructs are architectural components whose independent evidence is limited to the reported ablations.

free parameters (4)
  • epsilon (termination threshold) = 0.01
    Set by hand to 0.01; controls maximum iterations and therefore success/iteration trade-off.
  • actor temperature = 1.0
    Swept over {0.1,0.5,1.0,1.5,1.9}; final claims use 1.0.
  • alpha, beta of human-calibration sigmoid g = fitted on acceptance data
    Optimized by minimizing expected negative log acceptance on the training set; absolute values retained.
  • human-acceptance network widths = 10×24, 24×12, 12×1
    Chosen as 10-24-12-1; early-stopped with Adam.
axioms (4)
  • domain assumption An LLM evaluator’s binary correctness and agreement scores are sufficiently accurate proxies for ground-truth factual correctness and constraint satisfaction.
    Used throughout Sections 4–6 and the success definition; never independently validated against external oracles.
  • domain assumption A three-layer network trained on a re-balanced multi-task human-AI interaction dataset generalizes to tourism recommendations and to the demographic samples drawn at test time.
    Stated in Section 5 and Appendix A; underpins every reported acceptance probability.
  • ad hoc to paper The human loss L_T and the derived termination inequality (Eq. 1) correctly capture the utility a real user would experience.
    Introduced in Section 3.2; the theorem proves termination under this definition but not that the definition matches real utility.
  • standard math Standard stochastic-game and Reflexion memory constructions (short-term + long-term buffers) are valid.
    Imported from prior literature without modification.
invented entities (2)
  • Human-Centric Reflective Architecture (HCRA) no independent evidence
    purpose: Orchestrates actor, evaluator, self-reflection LLM, human acceptance model and human calibration model into a single iterative loop.
    The central architectural contribution; independent evidence is limited to the tourism ablations.
  • Human-calibrated confidence g_t no independent evidence
    purpose: Transforms raw AI confidence so that the acceptance model better predicts real-user behavior.
    Defined by the two-parameter sigmoid in Section 5; no external validation beyond the 2 % acceptance lift on the training distribution.

pith-pipeline@v1.1.0-grok45 · 27433 in / 3088 out tokens · 27544 ms · 2026-07-12T05:22:10.413311+00:00 · methodology

0 comments
read the original abstract

The use of Large Language Models (LLMs) across diverse areas of human activity-ranging from everyday tasks to safety-critical applications-aims to enhance decision-making effectiveness with minimal human feedback. Concurrently, it seeks to align decisions with human expectations, preferences, and needs while mitigating risks associated with AI non-determinism. However, humans frequently over- or under-rely on AI recommendations, and current AI systems remain poorly calibrated to human expectations. To address these challenges, we introduce a human-AI collaborative decision-making framework designed to augment human capabilities and align AI agents with human preferences and expectations. Specifically, this paper (a) formulates the collaborative decision-making task as a stochastic game between an AI agent and a human player, and (b) proposes the Human-Centric Reflective Architecture (HCRA), which integrates human-calibrated models with reinforcement learning agents that leverage linguistic feedback in an iterative, reflective process. Evaluation results demonstrate that HCRA enhances decision-making effectiveness and delivers high-quality recommendations.

Figures

Figures reproduced from arXiv: 2607.03025 by Andreas Kouridakis, Dimitrios Patiniotis Spyropoulos, George Vouros.

Figure 1
Figure 1. Figure 1: The overall human-centric reflective architecture. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Prompt templates [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of reflective text for an actor-generated [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The effect of (left) agreement on ℎ𝑎𝑐𝑐 and (right) of correctness on human-calibrated confidence 𝑔(𝑠𝑡 ). Experiments aim to show that: (a) HCRA-driven decision-making is more effective than a reflective agentic architecture without mod￾els of human behavior; (b) the human-centric aspects of HCRA play a significant role in effectiveness of the collaborative approach; (c) the proposed architecture facilitate… view at source ↗
Figure 6
Figure 6. Figure 6: Iterations required per question (actor temperature [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Iterations required for successful termination of [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Iterations required for termination when using [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 11
Figure 11. Figure 11 [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 9
Figure 9. Figure 9: Distribution of samples in tasks, in the original [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Distribution of samples in tasks, in our dataset [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Distribution of samples in geographical regions, [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 13
Figure 13. Figure 13: Distribution of total vs successful tasks (actor tem [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Iterations required per question (actor tempera [PITH_FULL_IMAGE:figures/full_fig_p017_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Iterations required for successful termination of [PITH_FULL_IMAGE:figures/full_fig_p017_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Distribution of total vs successful tasks (actor tem [PITH_FULL_IMAGE:figures/full_fig_p018_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Iterations required per question (actor tempera [PITH_FULL_IMAGE:figures/full_fig_p018_17.png] view at source ↗
Figure 21
Figure 21. Figure 21: Iterations required for successful termination of [PITH_FULL_IMAGE:figures/full_fig_p018_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Distribution of total vs successful tasks (actor tem [PITH_FULL_IMAGE:figures/full_fig_p019_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Iterations required per question (actor tempera [PITH_FULL_IMAGE:figures/full_fig_p019_23.png] view at source ↗
Figure 27
Figure 27. Figure 27: Iterations required for successful termination of [PITH_FULL_IMAGE:figures/full_fig_p019_27.png] view at source ↗
Figure 31
Figure 31. Figure 31: Unreliable evaluator: Distribution of total vs suc [PITH_FULL_IMAGE:figures/full_fig_p020_31.png] view at source ↗
Figure 29
Figure 29. Figure 29: No g: Iterations required per question (actor tem [PITH_FULL_IMAGE:figures/full_fig_p020_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: No g: Iterations required for successful termina [PITH_FULL_IMAGE:figures/full_fig_p020_30.png] view at source ↗
Figure 34
Figure 34. Figure 34: Prompt templates for the actor, evaluation and self-reflection models. [PITH_FULL_IMAGE:figures/full_fig_p021_34.png] view at source ↗
Figure 35
Figure 35. Figure 35: The reflective process for Question 24 [PITH_FULL_IMAGE:figures/full_fig_p022_35.png] view at source ↗
Figure 36
Figure 36. Figure 36: The reflective process for Question 16 [PITH_FULL_IMAGE:figures/full_fig_p023_36.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

31 extracted references · 2 canonical work pages

  1. [1]

    Gagan Bansal, Besmira Nushi, Ece Kamar, Eric Horvitz, and Daniel S. Weld

  2. [2]

    Proceedings of the AAAI Conference on Artificial Intelligence35, 13 (May 2021), 11405–11414

    Is the Most Accurate AI the Best Teammate? Optimizing AI for Teamwork. Proceedings of the AAAI Conference on Artificial Intelligence35, 13 (May 2021), 11405–11414. doi:10.1609/aaai.v35i13.17359

    doi:10.1609/aaai.v35i13.17359 2021

  3. [3]

    Lasecki, Daniel S

    Gagan Bansal, Besmira Nushi, Ece Kamar, Walter S. Lasecki, Daniel S. Weld, and Eric Horvitz. 2019. Beyond Accuracy: The Role of Mental Models in Human-AI Team Performance.Proceedings of the AAAI Conference on Human Computation and Crowdsourcing7, 1 (Oct. 2019), 2–11. doi:10.1609/hcomp.v7i1.5285 Andreas Kouridakis, Dimitrios Patiniotis Spyropoulos, and Geo...

    doi:10.1609/hcomp.v7i1.5285 2019

  4. [4]

    Gagan Bansal, Tongshuang Wu, Joyce Zhou, Raymond Fok, Besmira Nushi, Ece Kamar, Marco Tulio Ribeiro, and Daniel Weld. 2021. Does the Whole Exceed its Parts? The Effect of AI Explanations on Complementary Team Performance. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems (Yokohama, Japan)(CHI ’21). Association for Computing M...

    doi:10.1145/3411764.3445717 2021

  5. [5]

    Vardoulakis

    Emma Beede, Elizabeth Baylor, Fred Hersch, Anna Iurchenko, Lauren Wilcox, Paisan Ruamviboonsuk, and Laura M. Vardoulakis. 2020. A Human-Centered Evaluation of a Deep Learning System Deployed in Clinics for the Detection of Diabetic Retinopathy. InProceedings of the 2020 CHI Conference on Human Factors in Computing Systems(Honolulu, HI, USA)(CHI ’20). Asso...

    doi:10.1145/3313831.3376718 2020

  6. [6]

    Corvelo Benz and Manuel Gomez Rodriguez

    Nina L. Corvelo Benz and Manuel Gomez Rodriguez. 2024. Human-Aligned Calibration for AI-Assisted Decision Making. arXiv:2306.00074 [cs.LG] https: //arxiv.org/abs/2306.00074

    Pith/arXiv arXiv 2024

  7. [7]

    Jessica Y Bo, Sophia Wan, and Ashton Anderson. 2025. To Rely or Not to Rely? Evaluating Interventions for Appropriate Reliance on Large Language Models. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25). Association for Computing Machinery, New York, NY, USA, Article 905, 23 pages. doi:10.1145/3706598.3714097

    doi:10.1145/3706598.3714097 2025

  8. [8]

    Gajos, and Elena L

    Zana Buçinca, Phoebe Lin, Krzysztof Z. Gajos, and Elena L. Glassman. 2020. Proxy tasks and subjective measures can be misleading in evaluating explainable AI systems. InProceedings of the 25th International Conference on Intelligent User Interfaces(Cagliari, Italy)(IUI ’20). Association for Computing Machinery, New York, NY, USA, 454–464. doi:10.1145/3377...

    doi:10.1145/3377325.3377498 2020

  9. [9]

    Kevin Chen, Marco Cusumano-Towner, Brody Huval, Aleksei Petrenko, Jackson Hamburger, Vladlen Koltun, and Philipp Krähenbühl. 2025. Reinforcement Learning for Long-Horizon Interactive LLM Agents. arXiv:2502.01600 [cs.LG] https://arxiv.org/abs/2502.01600

    Pith/arXiv arXiv 2025

  10. [10]

    Unai Fischer-Abaigar, Christoph Kern, Noam Barda, and Frauke Kreuter. 2024. Bridging the gap: Towards an expanded toolkit for AI-driven decision-making in the public sector.Government Information Quarterly41, 4 (2024), 101976. doi:10.1016/j.giq.2024.101976

    work page doi:10.1016/j.giq.2024.101976 2024

  11. [11]

    Aaron Jaech and et al. 2024. OpenAI o1 System Card. arXiv:2412.16720 [cs.AI] https://arxiv.org/abs/2412.16720

    Pith/arXiv arXiv 2024

  12. [12]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. 2017. Adam: A Method for Stochastic Opti- mization. arXiv:1412.6980 [cs.LG] https://arxiv.org/abs/1412.6980

    Pith/arXiv arXiv 2017

  13. [13]

    Vera Liao, and Chenhao Tan

    Vivian Lai, Chacha Chen, Alison Smith-Renner, Q. Vera Liao, and Chenhao Tan

  14. [14]

    InProceedings of the 2023 ACM Con- ference on Fairness, Accountability, and Transparency(Chicago, IL, USA)(FAccT ’23)

    Towards a Science of Human-AI Decision Making: An Overview of Design Space in Empirical Human-Subject Studies. InProceedings of the 2023 ACM Con- ference on Fairness, Accountability, and Transparency(Chicago, IL, USA)(FAccT ’23). Association for Computing Machinery, New York, NY, USA, 1369–1385. doi:10.1145/3593013.3594087

    doi:10.1145/3593013.3594087 2023

  15. [15]

    Zhuoyan Li and Ming Yin. 2025. Utilizing human behavior modeling to manip- ulate explanations in AI-assisted decision making: the good, the bad, and the scary. InProceedings of the 38th International Conference on Neural Information Processing Systems(Vancouver, BC, Canada)(NIPS ’24). Curran Associates Inc., Red Hook, NY, USA, Article 163, 23 pages

    2025

  16. [16]

    Han Liu, Vivian Lai, and Chenhao Tan. 2021. Understanding the Effect of Out- of-distribution Examples and Interactive Explanations on Human-AI Decision Making.Proc. ACM Hum.-Comput. Interact.5, CSCW2, Article 408 (Oct. 2021), 45 pages. doi:10.1145/3479552

    doi:10.1145/3479552 2021

  17. [17]

    Shuai Ma, Ying Lei, Xinru Wang, Chengbo Zheng, Chuhan Shi, Ming Yin, and Xiaojuan Ma. 2023. Who Should I Trust: AI or Myself? Leveraging Human and AI Correctness Likelihood to Promote Appropriate Trust in AI-Assisted Decision- Making. InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems(Hamburg, Germany)(CHI ’23). Association for...

    doi:10.1145/3544548.3581058 2023

  18. [18]

    Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. 2022. Training language models to follow instructions with human fee...

    2022

  19. [19]

    Vouros, Ian Crook, Natalia Andrienko, Gennady Andrienko, and Jose Manuel Cordero

    George Papadopoulos, Alevizos Bastas, George A. Vouros, Ian Crook, Natalia Andrienko, Gennady Andrienko, and Jose Manuel Cordero. 2024. Deep reinforce- ment learning in service of air traffic controllers to resolve tactical conflicts.Ex- pert Systems with Applications236 (2024), 121234. doi:10.1016/j.eswa.2023.121234

    work page doi:10.1016/j.eswa.2023.121234 2024

  20. [20]

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: language agents with verbal reinforcement learn- ing. InProceedings of the 37th International Conference on Neural Information Processing Systems(New Orleans, LA, USA)(NIPS ’23). Curran Associates Inc., Red Hook, NY, USA, Article 377, 19 pages

    2023

  21. [21]

    Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2024. Scaling LLM Test- Time Compute Optimally can be More Effective than Scaling Model Parameters. arXiv:2408.03314 [cs.LG] https://arxiv.org/abs/2408.03314

    Pith/arXiv arXiv 2024

  22. [22]

    Takehiro Takayanagi, Ryuji Hashimoto, Chung-Chi Chen, and Kiyoshi Izumi

  23. [23]

    arXiv:2502.14311 [cs.HC] https://arxiv.org/abs/2502.14311

    The Impact and Feasibility of Self-Confidence Shaping for AI-Assisted Decision-Making. arXiv:2502.14311 [cs.HC] https://arxiv.org/abs/2502.14311

    Pith/arXiv arXiv

  24. [24]

    Bernstein, and Ranjay Krishna

    Helena Vasconcelos, Matthew Jörke, Madeleine Grunde-McLaughlin, Tobias Ger- stenberg, Michael S. Bernstein, and Ranjay Krishna. 2022. Explanations Can Re- duce Overreliance on AI Systems During Decision-Making.CoRRabs/2212.06823 (2022). arXiv:2212.06823 doi:10.48550/ARXIV.2212.06823

    Pith/arXiv arXiv doi:10.48550/arxiv.2212.06823 2022

  25. [25]

    Kailas Vodrahalli, Roxana Daneshjou, Tobias Gerstenberg, and James Zou. 2022. Do Humans Trust Advice More if it Comes from AI? An Analysis of Human- AI Interactions. InProceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society(Oxford, United Kingdom)(AIES ’22). Association for Computing Machinery, New York, NY, USA, 763–777. doi:10.1145/351409...

    doi:10.1145/3514094.3534150 2022

  26. [26]

    Kailas Vodrahalli, Tobias Gerstenberg, and James Zou. 2022. Uncalibrated models can improve Human-AI collaboration. InProceedings of the 36th International Conference on Neural Information Processing Systems(New Orleans, LA, USA) (NIPS ’22). Curran Associates Inc., Red Hook, NY, USA, Article 289, 13 pages

    2022

  27. [27]

    Xinru Wang and Ming Yin. 2021. Are Explanations Helpful? A Comparative Study of the Effects of Explanations in AI-Assisted Decision-Making. InProceedings of the 26th International Conference on Intelligent User Interfaces(College Station, TX, USA)(IUI ’21). Association for Computing Machinery, New York, NY, USA, 318–328. doi:10.1145/3397481.3450650

    doi:10.1145/3397481.3450650 2021

  28. [28]

    Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, Chenyang Shao, Yuwei Yan, Qinglong Yang, Yiwen Song, Sijian Ren, Xinyuan Hu, Yu Li, Jie Feng, Chen Gao, and Yong Li. 2025. Towards Large Reasoning Models: A Survey of Reinforced Reasoning with Large Language Models. arXiv:...

    Pith/arXiv arXiv 2025

  29. [29]

    Peter Zhang. 2023. Taking advice from chatgpt.arXiv preprint arXiv:2305.11888 (2023)

    Pith/arXiv arXiv 2023

  30. [30]

    Vera Liao, and Rachel K

    Yunfeng Zhang, Q. Vera Liao, and Rachel K. E. Bellamy. 2020. Effect of confidence and explanation on accuracy and trust calibration in AI-assisted decision making. InProceedings of the 2020 Conference on Fairness, Accountability, and Transparency (Barcelona, Spain)(FAT* ’20). Association for Computing Machinery, New York, NY, USA, 295–305. doi:10.1145/335...

    doi:10.1145/3351095.3372852 2020

  31. [31]

    sample" denotes a single inter- action comprising a question, an AI recommendation, and a person response. We consider samples of “agreement

    Daniel M. Ziegler, Nisan Stiennon, Jeffrey Wu, Tom B. Brown, Alec Radford, Dario Amodei, Paul Christiano, and Geoffrey Irving. 2020. Fine-Tuning Language Models from Human Preferences. arXiv:1909.08593 [cs.CL] https://arxiv.org/ abs/1909.08593 Human-Centric Reflective Architecture for Human-AI Collaborative Decision-Making APPENDIX A Dataset and Training ...

    Pith/arXiv arXiv 2020