Pith. sign in

REVIEW 3 major objections 5 minor 105 references

Reasoning Like an Economist: Post-Training on Economic Problems Induces Strategic Generalization in LLMs

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Economics post-training makes LLMs rational in unseen games.

desk verdict Plausible and interesting transfer result, but the missing control condition means the causal claim—economic content drives the game-playing gains—is not yet established. read the letter →

arxiv 2506.00577 v1 pith:25FCK7VR submitted 2025-05-31 cs.AI cs.CLcs.GTcs.MA

Yufa Zhou , Shaobo Wang , Xingyu Dong , Xiangqi Jin , Yifang Chen , Yue Min , Kexin Yang , Xingzhang Ren
show 2 more authors
Dayiheng Liu Linfeng Zhang
This is my paper
classification cs.AIcs.CLcs.GTcs.MA
keywords economicreasoningpost-trainingreinforcementlearningwithverifiablerewardsstrategicgeneralizationmulti-agentgamesgametheorygrouprelativepolicyoptimizationNashequilibrium
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

What the paper tries to establish is that a language model can become a better strategic player in multi-agent games without ever being trained on multi-agent interaction. The recipe is domain-aligned post-training: a 7B math-reasoning model is fine-tuned on 2,100 economic problems—backward induction, trigger strategies, equilibrium detection, behavioral biases—first by supervised imitation of distilled reasoning traces and then by reinforcement learning with a rule-based verifier. On two game benchmarks the model never saw, the trained model reaches a Nash equilibrium in self-play more often (0.59 to 0.685) and beats a strong fixed opponent more often (GTBench win rate 0.49 to 0.56). The paper interprets these gains as a transferable 'equilibrium prior': solving equilibrium problems offline instills a bias toward mutual best responses that carries over into live play. If true, this matters because direct multi-agent training is expensive and unstable, while single-agent problems with verifiable answers are cheap to produce at scale.

What carries the argument

The central mechanism is the two-stage post-training loop and the data that feeds it, packaged as Recon (Reasoning like an ECONomist). The Recon Dataset holds 2,100 multiple-choice economic problems across 15 categories, deliberately over-weighted toward game theory—enforceability, trigger strategies, and backward induction dominate—drawn from STEER, EconLogicQA, EconNLI, and pure-strategy equilibrium games. Stage one is supervised fine-tuning on 868 chains of thought distilled from a stronger teacher (QwQ-32B), standardized so reasoning sits inside <think>...</think> and the verdict in \boxed{}. Stage two is Group Relative Policy Optimization (GRPO) with a hierarchical rule-based reward (+5 exact match, +2 partial, -3 incorrect, -4 format violation, -5 parse failure). The proposed transfer mechanism is an 'equilibrium prior': because GRPO's advantage rewards final correctness, the model learns to plan backward—selecting steps that survive any continuation—so repeated practice on equilibrium problems biases it toward mutual best responses when it later faces an opponent; a secondary channel is format transfer, since the think-then-commit interface mirrors the inner-rollout/outer-commitment loop of live game play.

What would settle it

Run the identical SFT-to-GRPO pipeline on a matched set of non-economic reasoning problems with the same format, counts, and reward schedule—for example, math word problems or logic puzzles with verifiable answers—and evaluate on the same ten complete-information games and ten GTBench tasks. If the control reaches Recon-RL's 0.685 self-play Nash frequency and 0.56 win rate, the economic content is not the causal driver; if it stays near the 0.59 and 0.49 baselines, the economic content carries the effect.

Watch

Extended reading notes

Core claim

The paper's central claim is that post-training on economic reasoning induces strategic generalization: models trained on economic problems display economically rational behavior in multi-agent games despite receiving no interaction-based supervision (Abstract). Concretely, the Recon pipeline—supervised fine-tuning on 868 teacher-distilled chains of thought, then Group Relative Policy Optimization against a hierarchical rule-based reward—takes DeepSeek-R1-Distill-Qwen-7B from 0.59 to 0.685 in self-play Nash-equilibrium frequency across ten complete-information games, and from 0.49 to 0.56 in GTBench win rate against GPT-4o-mini, nearly double the 0.345 of the non-reasoning Qwen2.5-7B-Instruct. The largest relative gains appear in negotiation and breakthrough, games demanding extended look-ahead and adaptive bidding that never appeared in the training corpus. The paper takes the equilibrium-rate jump as quantitative evidence of a transferable equilibrium prior, and treats the unchanged score on nim—whose solution is a single XOR invariant—as the boundary showing that the training injects an equilibrium-seeking bias rather than a generic combinatorial search skill.

Load-bearing premise

The paper's conclusion depends on the assumption that the game-playing gains come from the economic content of its training data rather than from generic improvements any post-training on structured reasoning would produce—an assumption untested because Section 5.3 compares only against the untrained base model and a non-reasoning instruct model, never a model post-trained on non-economic reasoning.

Editorial extensions

If this is right

  • Interaction-free alignment: verifiable single-agent problems can substitute for interaction-based supervision in producing strategically competent agents.
  • A division of labor between stages: SFT installs the formal vocabulary and structure of game-theoretic reasoning, while GRPO teaches the search procedure (tree construction, backward induction, self-correction) visible in the qualitative traces.
  • Accessibility: the effect appears at 7B scale on a single H800 GPU, so domain-aligned post-training is a practical route to strategic behavior rather than a large-scale enterprise.
  • The boundary of the effect: performance on nim is unchanged, so the injected bias is equilibrium-seeking, not a general combinatorial problem solver; gains should concentrate in incentive-driven games.
  • Benchmark-side expectations: the largest transfers are to negotiation (+0.20) and breakthrough (+0.20), games that demand extended look-ahead and adaptive bidding absent from the training data.

Reading between the lines

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

  • The paper does not train a control on matched non-economic reasoning problems, so its attribution of the gains to economics specifically rather than to generic SFT/RL improvement is untested; a control on math or logic problems at equal scale and format would settle whether the equilibrium prior is economic or just an artifact of structured-reasoning post-training.
  • The training data is heavily weighted toward repeated-game and extensive-form concepts (enforceability, trigger, and backward induction are roughly 42 percent of items), so the transfer might hinge on that specific distribution; reweighting toward behavioral-economics categories such as endowment and certainty effects should shrink the game-play gains if game-theory content is the active ingredien
  • The think-then-commit template is itself a candidate mechanism: if the format enforces an inner-rollout/outer-commitment loop, then identical content trained without <think> markers should transfer less, a directly testable ablation.
  • If the equilibrium prior is real, it should show up in mixed-strategy and Bayesian settings, such as moves matching the mixed equilibrium of matching pennies or bid shading in private-value auctions, which the paper's evaluation games do not fully exercise.
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 introduces Recon, a 7B LLM post-trained via Supervised Fine-Tuning (SFT) and Group Relative Policy Optimization (GRPO) on a hand-curated dataset of 2,100 economic reasoning problems drawn from STEER, EconLogicQA, EconNLI, and a pure-strategy equilibrium dataset. Reasoning traces are distilled from QwQ-32B to form the Recon-CoT corpus used for SFT, and a hierarchical rule-based reward is used for GRPO. The resulting Recon-SFT and Recon-RL models are evaluated on a held-out economic reasoning set (Recon-Eval) and on two multi-agent game suites: self-play on the Complete-Information Games framework, measured by Nash equilibrium frequency, and GTBench against GPT-4o-mini, measured by win rate. The paper reports a 14.7% absolute accuracy gain on Recon-Eval, an increase in Nash equilibrium frequency from 0.59 to 0.685, and an increase in GTBench win rate from 0.49 to 0.56, interpreting these as evidence that economic post-training induces strategic generalization to unseen multi-agent games. Qualitative trace analysis on the Draco game is presented as evidence of emergent strategic modeling after RL.

Significance. If the causal claim were established, the paper would offer a scalable and low-cost route toward aligning LLM agents with game-theoretic rationality, potentially complementing multi-agent co-training and interaction-based supervision. The work has tangible strengths: the code is released, the reward function is rule-based and transparent, the evaluations use publicly available benchmarks (GTBench, Complete-Information Games), and the paper distinguishes between SFT and RL contributions to the observed gains. However, the central claim that economic content, rather than generic post-training, drives the strategic improvements is not yet supported by the experimental design, and several quantitative results are presented without uncertainty quantification. With a matched non-economic control and more rigorous statistical reporting, the contribution would be a useful step toward understanding when and how single-agent post-training transfers to multi-agent settings.

major comments (3)
  1. [§5.3, Tables 4–5] The abstract's causal claim that post-training on economic problems 'induces strategic generalization' is not established without a control for the post-training procedure itself. The comparisons are only against the untrained DeepSeek-R1-Distill-Qwen-7B and Qwen2.5-7B-Instruct, so the observed gains in Nash equilibrium frequency (0.59→0.685) and GTBench win rate (0.49→0.56) could be attributable to generic SFT/GRPO effects—improved instruction following, format adherence, longer chain-of-thought, or self-correction—rather than to the economic content of Recon. The unchanged nim score in Table 5 rules out memorization of a specific combinatorial trick but does not rule out these generic mechanisms. Moreover, because the Recon training set includes a large share of game-theoretic material (Table 6 lists Enforceability, Backward Induction, Trigger, Feasibility, Bayes Nash, Pure Nash, and PTE, totaling roughly 1,150 of 1,800 items), the transfer from 'economic problems' to game play may be near-domain transfer rather than a general economic-to-strategic alignment effect. A matched non-economic control model trained with the identical SFT+GRPO pipeline on, for example, mathematics or general reasoning, or an ablation that removes game-theoretic categories, is required to support the causal interpretation.
  2. [§5.1, Tables 4–5] The multi-agent evaluations report 20 trials per Complete-Information game and 10 trials per GTBench task, but no error bars, confidence intervals, or significance tests are provided. The apparent improvements are often small relative to sampling noise: in Table 5, the R1-Distill-to-SFT overall win-rate increase is 0.04 (49 vs 53 wins across 100 trials, approximate standard error 0.05), and per-game changes of 0.1 correspond to a single trial. Without uncertainty quantification, statements such as 'GRPO again provides a consistent lift' (Section 5.3) are not supported. The authors should report bootstrap or exact binomial confidence intervals and, ideally, paired comparisons across games.
  3. [§4.3 and §5.2, Table 3] Recon-Eval is described as a 'held-out' set, but it is constructed from the same 15 categories and the same underlying sources (STEER, EconLogicQA, EconNLI, Pure-Strategy Equilibrium Games) as the training split, with 20 questions per category mirroring the training proportions. Consequently, the 14.7% absolute gain (48.3%→63.0%) is an in-distribution accuracy improvement rather than evidence of generalization to new economic reasoning domains. The authors should either qualify 'held-out' to mean 'held-out within the same distribution' or add an evaluation on categories not seen in training; the current wording in the abstract and conclusion overstates the result.
minor comments (5)
  1. [Table 2] There are typos in the column headers: 'Enviornments' should be 'Environments', and abbreviations such as 'Dom’d' are not defined in the table or caption.
  2. [Figure 1] The pipeline diagram is dense and difficult to read in the provided version; consider enlarging the figure or splitting it into two panels for clarity.
  3. [§6.1] The two proposed mechanisms ('structured prompts ⇒ modular latent policies' and 'outcome-aligned reward ⇒ an equilibrium prior') are presented in the main text as explanations of the results, but they are conjectures; the authors should label them explicitly as hypotheses rather than conclusions.
  4. [Abstract and Conclusion] The phrase 'isolated economic benchmarks' in the conclusion is ambiguous; 'single-agent benchmarks' would be clearer.
  5. [General] The paper lacks a limitations section; the absence of a non-economic control, the small number of game trials, and the in-distribution nature of Recon-Eval should be acknowledged explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the strategic-generalization claim is tested on unseen games against the same base model, and no load-bearing step reduces to its inputs.

full rationale

The central claim, that post-training on economic problems induces strategic generalization, is supported by comparisons on external, unseen interactive benchmarks (Complete-Information Games [29] and GTBench [17], Tables 4-5) between the same base model before and after SFT/GRPO. These evaluations do not reduce to training inputs by construction: the games are not part of the Recon corpus, the GRPO rewards are correctness on economic multiple-choice items (Section 3.5 and A.2), and no test-game quantity enters the training signal. The held-out Recon-Eval (20 items per category, mirroring the training distribution) is an in-distribution accuracy report, not a prediction that is equivalent to its inputs; selecting categories based on baseline error rates affects external validity but is not circular. The Related Work appendix contains several self-citations (e.g., Refs. [44], [45], [65]-[67], [76]-[80], [84], [91]), but none is load-bearing for the generalization claim, and no uniqueness theorem or ansatz is imported from prior work. The main weakness is the absence of a matched non-economic SFT/RL control, which leaves open whether generic post-training effects, rather than economic content, drive the gains; that is an experimental-design confound, not a circular derivation. No step in the derivation chain is equivalent to its own inputs.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central empirical claim does not rest on a closed-form derivation, so the main 'free parameters' are hand-chosen training and data-curation decisions. The key assumptions are the validity of benchmark labels and evaluation metrics, and the fairness of the unseen-game tests.

free parameters (3)
  • Reward values in the rule-based GRPO reward function = +5 exact, +2 partial, -3 incorrect, -4 format violation, -5 parse failure
    Hand-chosen in Section A.2; no ablation is provided to show these specific values are optimal or that the generalization result is robust to them.
  • Recon dataset category quotas and category selection = e.g., 250 Enforceability, 250 Backward Induction, 250 Trigger, 150 Feasibility, 75 Endowment Effect, etc. per Table 6
    The 15 categories were selected based on baseline error rates and sampled proportionally to those errors (Section 4.3); this hand-crafted distribution shapes both training and the in-domain evaluation set.
  • Training hyperparameters = SFT: lr 2e-4, batch size 8, 2700 steps, LoRA rank 8; RL: lr 5e-6, batch 32, 8 samples, 2250 steps
    Reported in Section A.4 without sensitivity analysis; these choices affect the magnitude of gains but are standard for the base model and framework.
assumptions (4)
  • domain assumption Benchmark labels in STEER, EconLogicQA, EconNLI, and the pure-strategy game dataset are correct and represent economic rationality.
    Both SFT (correctness filtering) and RL rewards (Section A.2) rely on these labels as ground truth; systematic mislabeling would propagate into training and evaluation.
  • domain assumption Nash equilibrium frequency in self-play is a valid proxy for economic rationality and strategic competence.
    Section 5.1 uses Nash equilibrium frequency as the primary self-play metric; this assumes that equilibrium convergence is desirable and that the metric captures strategic skill rather than, say, coordination artifacts.
  • domain assumption The evaluation games (Complete-Information Games and GTBench) were not seen during post-training and are a fair test of generalization.
    The paper states no interaction-based supervision was used (Section 5.3), but it does not report a contamination check, so leakage through pretraining or distillation cannot be fully ruled out.
  • domain assumption QwQ-32B's reasoning traces that end in the correct answer are high-quality demonstrations for SFT.
    Section 4.4 filters traces only by final-answer correctness; a correct answer can accompany flawed reasoning, and SFT would imitate that reasoning.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reasoning Like an Economist: Post-Training on Economic Problems Induces Strategic Generalization in LLMs." pith.science (2026). https://pith.science/paper/25FCK7VR

@misc{pith2026250600577,
  author       = {Pith},
  title        = {Pith review of: Reasoning Like an Economist: Post-Training on Economic Problems Induces Strategic Generalization in LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/25FCK7VR}},
  note         = {Machine review of arXiv:2506.00577}
}
abstract

Directly training Large Language Models (LLMs) for Multi-Agent Systems (MAS) remains challenging due to intricate reward modeling, dynamic agent interactions, and demanding generalization requirements. This paper explores whether post-training techniques, specifically Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Rewards (RLVR), can effectively $\textit{generalize}$ to multi-agent scenarios. We use economic reasoning as a testbed, leveraging its strong foundations in mathematics and game theory, its demand for structured analytical reasoning, and its relevance to real-world applications such as market design, resource allocation, and policy analysis. We introduce $\textbf{Recon}$ ($\textbf{R}$easoning like an $\textbf{ECON}$omist), a 7B-parameter open-source LLM post-trained on a hand-curated dataset of 2,100 high-quality economic reasoning problems. Comprehensive evaluation on economic reasoning benchmarks and multi-agent games reveals clear improvements in structured reasoning and economic rationality. These results underscore the promise of domain-aligned post-training for enhancing reasoning and agent alignment, shedding light on the roles of SFT and RL in shaping model behavior. Code is available at https://github.com/MasterZhou1/Recon .

Figures

Figures reproduced from arXiv: 2506.00577 by the authors.

Figure 1
Figure 1. Overview of the Recon Pipeline. Step 1: We curate a high-quality economic dataset (Recon Dataset) from benchmarks such as STEER, and distill reasoning traces from teacher models to construct the Recon-CoT Dataset. Step 2: A base model is post-trained via supervised fine-tuning (Recon-SFT) on Recon-CoT and reinforcement learning (Recon-RL) on the Recon Dataset. Step 3: The resulting models are evaluated on reasoning … view at source ↗
Figure 2
Figure 2. Training dynamics for SFT (a) and RL (b). [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Example question prompt for backward_induction used in the Dataset Curation Experiment. Example Recon Training Dataset Question Prompt You are given a reasoning problem in Economics. Please reason step by step to solve it. Then, give your final answer using the format \boxed{Option X: full choice text}. You must include both the option number and the full text of your chosen answer. Question: Consider a duopoly wher… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Example question prompt used in the Recon training dataset. [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: Example Draco game [29] output of Recon-RL. Some important game theory terminology employed by the model is bolded for reference. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]
Figure 6
Figure 6. Figure 6: Example Draco game [29] output of Recon-SFT. Some important game theory terminology employed by the model is bolded for reference. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

105 extracted references · 26 canonical work pages

  1. [1]

    Coop- eration, competition, and maliciousness: LLM-stakeholders interactive negotiation

    Sahar Abdelnabi, Amr Gomaa, Sarath Sivaprasad, Lea Schönherr, and Mario Fritz. Coop- eration, competition, and maliciousness: LLM-stakeholders interactive negotiation. InThe Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2024. URLhttps://openreview.net/forum?id=59E19c6yrN

    2024

  2. [2]

    Playing repeated games with large language models.Nature Human Behaviour, pages 1–11, 2025

    Elif Akata, Lion Schulz, Julian Coda-Forno, Seong Joon Oh, Matthias Bethge, and Eric Schulz. Playing repeated games with large language models.Nature Human Behaviour, pages 1–11, 2025. doi: 10.1038/s41562-025-02172-y. URL https://doi.org/10.1038/ s41562-025-02172-y

    doi:10.1038/s41562-025-02172-y 2025

  3. [3]

    Mechanistic interpretability for AI safety - a review

    Leonard Bereska and Stratis Gavves. Mechanistic interpretability for AI safety - a review. Transactions on Machine Learning Research, 2024. ISSN 2835-8856. URL https:// openreview.net/forum?id=ePUVetPKu6. Survey Certification, Expert Certification

    2024

  4. [4]

    Token merging: Your vit but faster.arXiv preprint arXiv:2210.09461, 2022

    Daniel Bolya, Cheng-Yang Fu, Xiaoliang Dai, Peizhao Zhang, Christoph Feichtenhofer, and Judy Hoffman. Token merging: Your vit but faster.arXiv preprint arXiv:2210.09461, 2022

    arXiv 2022

  5. [5]

    Sparks of artificial general intelligence: Early experiments with gpt-4, 2023

    Sébastien Bubeck, Varun Chadrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, et al. Sparks of artificial general intelligence: Early experiments with gpt-4, 2023

    2023

  6. [6]

    Cambridge University Press, 2006

    Nicolo Cesa-Bianchi and Gábor Lugosi.Prediction, Learning, and Games. Cambridge University Press, 2006

    2006

  7. [7]

    Towards reasoning era: A survey of long chain-of-thought for reasoning large language models.arXiv preprint arXiv:2503.09567, 2025

    Qiguang Chen, Libo Qin, Jinhao Liu, Dengyun Peng, Jiannan Guan, Peng Wang, Mengkang Hu, Yuhang Zhou, Te Gao, and Wanxiang Che. Towards reasoning era: A survey of long chain-of-thought for reasoning large language models.arXiv preprint arXiv:2503.09567, 2025

    arXiv 2025

  8. [8]

    The computational limits of state-space models and mamba via the lens of circuit complexity

    Yifang Chen, Xiaoyu Li, Yingyu Liang, Zhenmei Shi, and Zhao Song. The computational limits of state-space models and mamba via the lens of circuit complexity. InThe Second Conference on Parsimony and Learning (Proceedings Track), 2025. URL https://openreview.net/ forum?id=bImlLT3r62

    2025

Show all 105 references
  1. [9]

    Universal approximation of visual autoregressive transformers.arXiv preprint arXiv:2502.06167, 2025

    Yifang Chen, Xiaoyu Li, Yingyu Liang, Zhenmei Shi, and Zhao Song. Universal approximation of visual autoregressive transformers.arXiv preprint arXiv:2502.06167, 2025

    2025 arXiv

  2. [10]

    Sft memorizes, rl generalizes: A comparative study of foundation model post-training.arXiv preprint arXiv:2501.17161, 2025

    Tianzhe Chu, Yuexiang Zhai, Jihan Yang, Shengbang Tong, Saining Xie, Dale Schuurmans, Quoc V Le, Sergey Levine, and Yi Ma. Sft memorizes, rl generalizes: A comparative study of foundation model post-training.arXiv preprint arXiv:2501.17161, 2025

    2025 arXiv

  3. [11]

    Gamebench: Evaluating strategic reasoning abilities of LLM agents

    Anthony Costarelli, Mat Allen, Roman Hauksson, Grace Sodunke, Suhas Hariharan, Carlson Cheng, Wenjie Li, Joshua M Clymer, and Arjun Yadav. Gamebench: Evaluating strategic reasoning abilities of LLM agents. InLanguage Gamification - NeurIPS 2024 Workshop, 2024. URLhttps://openr...

    2024

  4. [12]

    Learning to estimate shapley values with vision transformers.arXiv preprint arXiv:2206.05282, 2022

    Ian Covert, Chanwoo Kim, and Su-In Lee. Learning to estimate shapley values with vision transformers.arXiv preprint arXiv:2206.05282, 2022. 10

    2022 arXiv

  5. [13]

    Unsloth, 2023

    Michael Han Daniel Han and Unsloth team. Unsloth, 2023. URL http://github.com/ unslothai/unsloth

    2023

  6. [14]

    Evaluating language model agency through negotiations

    Tim Ruben Davidson, Veniamin Veselovsky, Michal Kosinski, and Robert West. Evaluating language model agency through negotiations. InThe Twelfth International Conference on Learning Representations, 2024. URLhttps://openreview.net/forum?id=3ZqKxMHcAg

    2024

  7. [15]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    DeepSeek-AI. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    2025 arXiv

  8. [16]

    A survey on the optimization of large language model-based agents.arXiv preprint arXiv:2503.12434, 2025

    Shangheng Du, Jiabao Zhao, Jinxin Shi, Zhentao Xie, Xin Jiang, Yanhong Bai, and Liang He. A survey on the optimization of large language model-based agents.arXiv preprint arXiv:2503.12434, 2025

    2025

  9. [17]

    GTBench: Uncovering the strategic reasoning capabilities of LLMs via game-theoretic evaluations

    Jinhao Duan, Renming Zhang, James Diffenderfer, Bhavya Kailkhura, Lichao Sun, Elias Stengel-Eskin, Mohit Bansal, Tianlong Chen, and Kaidi Xu. GTBench: Uncovering the strategic reasoning capabilities of LLMs via game-theoretic evaluations. InThe Thirty- eighth Annual Conference...

    2024

  10. [18]

    Can large language models serve as rational players in game theory? a systematic analysis.Proceedings of the AAAI Conference on Artificial Intelligence, 38(16):17960–17967, Mar

    Caoyun Fan, Jindou Chen, Yaohui Jin, and Hao He. Can large language models serve as rational players in game theory? a systematic analysis.Proceedings of the AAAI Conference on Artificial Intelligence, 38(16):17960–17967, Mar. 2024. doi: 10.1609/aaai.v38i16.29751. URLhttps://o...

    2024 doi

  11. [19]

    How far are we from agi: Are llms all we need?Transactions on Machine Learning Research, 2024

    Tao Feng, Chuanyang Jin, Jingyu Liu, Kunlun Zhu, Haoqin Tu, Zirui Cheng, Guanyu Lin, and Jiaxuan You. How far are we from agi: Are llms all we need?Transactions on Machine Learning Research, 2024

    2024

  12. [20]

    Dataset with 200 million 3-by-3 strategic games for comparing perfectly transparent equilibria with nash equilibria, 2020-10-07

    Ghislain Fourny and Felipe Sulser Larraz. Dataset with 200 million 3-by-3 strategic games for comparing perfectly transparent equilibria with nash equilibria, 2020-10-07

    2020

  13. [21]

    Cognitive behaviors that enable self-improving reasoners, or, four habits of highly effective stars.arXiv preprint arXiv:2503.01307, 2025

    Kanishk Gandhi, Ayush Chakravarthy, Anikait Singh, Nathan Lile, and Noah D Goodman. Cognitive behaviors that enable self-improving reasoners, or, four habits of highly effective stars.arXiv preprint arXiv:2503.01307, 2025

    2025 arXiv

  14. [22]

    The llama 3 herd of models.arXiv e-prints, pages arXiv–2407, 2024

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models.arXiv e-prints, pages arXiv–2407, 2024

    2024

  15. [23]

    Econnli: Evaluating large language models on economics reasoning

    Yue Guo and Yi Yang. Econnli: Evaluating large language models on economics reasoning. InFindings of the Association for Computational Linguistics ACL 2024, pages 982–994, 2024

    2024

  16. [24]

    To- wards lossless dataset distillation via difficulty-aligned trajectory matching

    Ziyao Guo, Kai Wang, George Cazenavette, Hui Li, Kaipeng Zhang, and Yang You. To- wards lossless dataset distillation via difficulty-aligned trajectory matching. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  17. [25]

    A multi-llm-agent-based framework for economic and public policy analysis.arXiv preprint arXiv:2502.16879, 2025

    Yuzhi Hao and Danyang Xie. A multi-llm-agent-based framework for economic and public policy analysis.arXiv preprint arXiv:2502.16879, 2025

    2025

  18. [26]

    Measuring mathematical problem solving with the MATH dataset

    Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the MATH dataset. InThirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (...

    2021

  19. [27]

    Training compute-optimal large language models.arXiv preprint arXiv:2203.15556, 2022

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. Training compute-optimal large language models.arXiv preprint arXiv:2203.15556, 2022

    2022 arXiv

  20. [28]

    Lora: Low-rank adaptation of large language models.ICLR, 1 (2):3, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.ICLR, 1 (2):3, 2022. 11

    2022

  21. [29]

    Game-theoretic llm: Agent workflow for negotiation games.arXiv preprint arXiv:2411.05990, 2024

    Wenyue Hua, Ollie Liu, Lingyao Li, Alfonso Amayuelas, Julie Chen, Lucas Jiang, Mingyu Jin, Lizhou Fan, Fei Sun, William Wang, et al. Game-theoretic llm: Agent workflow for negotiation games.arXiv preprint arXiv:2411.05990, 2024

    2024 arXiv

  22. [30]

    Scaling laws for neural language models.arXiv preprint arXiv:2001.08361, 2020

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models.arXiv preprint arXiv:2001.08361, 2020

    2001 arXiv

  23. [31]

    Large language models are zero-shot reasoners.Advances in neural information processing systems, 35:22199–22213, 2022

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large language models are zero-shot reasoners.Advances in neural information processing systems, 35:22199–22213, 2022

    2022

  24. [32]

    Scaling laws for precision.arXiv preprint arXiv:2411.04330, 2024

    Tanishq Kumar, Zachary Ankner, Benjamin F Spector, Blake Bordelon, Niklas Muennighoff, Mansheej Paul, Cengiz Pehlevan, Christopher Ré, and Aditi Raghunathan. Scaling laws for precision.arXiv preprint arXiv:2411.04330, 2024

    2024 arXiv

  25. [33]

    Gonzalez, Hao Zhang, and Ion Stoica

    Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. Efficient memory management for large language model serving with pagedattention. InProceedings of the ACM SIGOPS 29th Symposium on Operating Systems ...

    2023

  26. [34]

    Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, Yuling Gu, Saumya Malik, Victoria Graf, Jena D

    Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V . Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, Yuling Gu, Saumya Malik, Victoria Graf, Jena D. Hwang, Jiangjiang Yang, Ronan Le Bras, Oyvind Tafjord, Chris Wilhelm, Lu...

    2025 arXiv

  27. [35]

    Investigating tax evasion emergence using dual large language model and deep reinforcement learning powered agent-based simulation.arXiv preprint arXiv:2501.18177, 2025

    Teddy Lazebnik and Labib Shami. Investigating tax evasion emergence using dual large language model and deep reinforcement learning powered agent-based simulation.arXiv preprint arXiv:2501.18177, 2025

    2025 arXiv

  28. [36]

    CAMEL: Communicative agents for "mind" exploration of large language model society

    Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. CAMEL: Communicative agents for "mind" exploration of large language model society. InAdvances in Neural Information Processing Systems, volume 36, pages 51991– 52008, 2023

    2023

  29. [37]

    EconAgent: Large language model-empowered agents for simulating macroeconomic activities

    Nian Li, Chen Gao, Mingyu Li, Yong Li, and Qingmin Liao. EconAgent: Large language model-empowered agents for simulating macroeconomic activities. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15523–155...

    2024

  30. [38]

    Limr: Less is more for rl scaling.arXiv preprint arXiv:2502.11886, 2025

    Xuefeng Li, Haoyang Zou, and Pengfei Liu. Limr: Less is more for rl scaling.arXiv preprint arXiv:2502.11886, 2025

    2025 arXiv

  31. [39]

    Beyond linear approximations: A novel pruning approach for attention matrix

    Yingyu Liang, Jiangxuan Long, Zhenmei Shi, Zhao Song, and Yufa Zhou. Beyond linear approximations: A novel pruning approach for attention matrix. InThe Thirteenth International Conference on Learning Representations, 2025. URL https://openreview.net/forum? id=sgbI8Pxwie

    2025

  32. [40]

    Looped relu mlps may be all you need as programmable computers

    Yingyu Liang, Zhizhou Sha, Zhenmei Shi, Zhao Song, and Yufa Zhou. Looped relu mlps may be all you need as programmable computers. InThe 28th International Conference on Artificial Intelligence and Statistics, 2025

    2025

  33. [41]

    Marft: Multi-agent reinforcement fine-tuning.arXiv preprint arXiv:2504.16129, 2025

    Junwei Liao, Muning Wen, Jun Wang, and Weinan Zhang. Marft: Multi-agent reinforcement fine-tuning.arXiv preprint arXiv:2504.16129, 2025

    2025 arXiv

  34. [42]

    Let’s verify step by step

    Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. Let’s verify step by step. InThe Twelfth International Conference on Learning Representations, 2023. 12

    2023

  35. [43]

    Awq: Activation-aware weight quanti- zation for on-device llm compression and acceleration.Proceedings of Machine Learning and Systems, 6:87–100, 2024

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei-Chen Wang, Guangx- uan Xiao, Xingyu Dang, Chuang Gan, and Song Han. Awq: Activation-aware weight quanti- zation for on-device llm compression and acceleration.Proceedings of Machine Learning and Systems, 6:87–100, 2024

    2024

  36. [44]

    Multi-stage vision token dropping: Towards efficient multimodal large language model.arXiv preprint arXiv:2411.10803, 2024

    Ting Liu, Liangtao Shi, Richang Hong, Yue Hu, Quanjun Yin, and Linfeng Zhang. Multi-stage vision token dropping: Towards efficient multimodal large language model.arXiv preprint arXiv:2411.10803, 2024

    2024 arXiv

  37. [45]

    Shifting ai efficiency from model-centric to data-centric compression.arXiv preprint arXiv:2505.19147, 2025

    Xuyang Liu, Zichen Wen, Shaobo Wang, Junjie Chen, Zhishan Tao, Yubo Wang, Xiangqi Jin, Chang Zou, Yiyu Wang, Chenfei Liao, et al. Shifting ai efficiency from model-centric to data-centric compression.arXiv preprint arXiv:2505.19147, 2025

    2025

  38. [46]

    Fin-r1: A large language model for financial reasoning through reinforcement learning.arXiv preprint arXiv:2503.16252, 2025

    Zhaowei Liu, Xin Guo, Fangqi Lou, Lingfeng Zeng, Jinyi Niu, Zixuan Wang, Jiajie Xu, Weige Cai, Ziwei Yang, Xueqian Zhao, et al. Fin-r1: A large language model for financial reasoning through reinforcement learning.arXiv preprint arXiv:2503.16252, 2025

    2025

  39. [47]

    Understanding r1-zero-like training: A critical perspective.arXiv preprint arXiv:2503.20783, 2025

    Zichen Liu, Changyu Chen, Wenjun Li, Penghui Qi, Tianyu Pang, Chao Du, Wee Sun Lee, and Min Lin. Understanding r1-zero-like training: A critical perspective.arXiv preprint arXiv:2503.20783, 2025

    2025 arXiv

  40. [48]

    Inference-time scaling for generalist reward modeling.arXiv preprint arXiv:2504.02495, 2025

    Zijun Liu, Peiyi Wang, Runxin Xu, Shirong Ma, Chong Ruan, Peng Li, Yang Liu, and Yu Wu. Inference-time scaling for generalist reward modeling.arXiv preprint arXiv:2504.02495, 2025

    2025

  41. [49]

    The llama 4 herd: The beginning of a new era of natively multimodal ai innovation

    AI @ Meta Llama Team. The llama 4 herd: The beginning of a new era of natively multimodal ai innovation. https://ai.meta.com/blog/llama-4-multimodal-intelligence/ , 2025. Accessed: April 5

    2025

  42. [50]

    Sql-r1: Training natural language to sql reasoning model by reinforcement learning.arXiv preprint arXiv:2504.08600, 2025

    Peixian Ma, Xialie Zhuang, Chengjin Xu, Xuhui Jiang, Ran Chen, and Jian Guo. Sql-r1: Training natural language to sql reasoning model by reinforcement learning.arXiv preprint arXiv:2504.08600, 2025

    2025

  43. [51]

    American invitational mathematics examination - aime

    MAA. American invitational mathematics examination - aime. InAmerican Invitational Mathematics Examination - AIME 2024, February 2024, 2024. URL https://maa.org/ math-competitions/american-invitational-mathematics-examination-aime

    2024

  44. [52]

    Tractable multi-agent reinforcement learning through behavioral economics

    Eric Mazumdar, Kishan Panaganti, and Laixi Shi. Tractable multi-agent reinforcement learning through behavioral economics. InThe Thirteenth International Conference on Learning Representations, 2025. URLhttps://openreview.net/forum?id=stUKwWBuBm

    2025

  45. [53]

    s1: Simple test-time scaling.arXiv preprint arXiv:2501.19393, 2025

    Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Candès, and Tatsunori Hashimoto. s1: Simple test-time scaling.arXiv preprint arXiv:2501.19393, 2025

    2025 arXiv

  46. [54]

    Introducing chatgpt

    OpenAI. Introducing chatgpt. https://openai.com/index/chatgpt/, 2022. Accessed: November 30

    2022

  47. [55]

    Hello gpt-4o

    OpenAI. Hello gpt-4o. https://openai.com/index/hello-gpt-4o/, 2024. Accessed: May 14

    2024

  48. [56]

    Openai o1 system card.arXiv preprint arXiv:2412.16720, 2024

    OpenAI. Openai o1 system card.arXiv preprint arXiv:2412.16720, 2024

    2024 arXiv

  49. [57]

    Long Ouyang, Jeffrey 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 Welin- der, Paul F. Christiano, Jan Le...

    2022

  50. [58]

    O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S

    Joon Sung Park, Joseph C. O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior. In Sean Follmer, Jeff Han, Jürgen Steimle, and Nathalie Henry Riche, editors,Proceedings of the 36th An...

    2023

  51. [59]

    Corrupted by reasoning: Reasoning language models become free-riders in public goods games

    David Guzman Piedrahita, Yongjin Yang, Mrinmaya Sachan, Giorgia Ramponi, Bernhard Schölkopf, and Zhijing Jin. Corrupted by reasoning: Reasoning language models become free-riders in public goods games. 2025. URL https://zhijing-jin.com/files/papers/ 2025_SanctSim.pdf

    2025

  52. [60]

    Fino1: On the transferability of reasoning enhanced llms to finance.arXiv preprint arXiv:2502.08127, 2025

    Lingfei Qian, Weipeng Zhou, Yan Wang, Xueqing Peng, Han Yi, Jimin Huang, Qianqian Xie, and Jianyun Nie. Fino1: On the transferability of reasoning enhanced llms to finance.arXiv preprint arXiv:2502.08127, 2025

    2025 arXiv

  53. [61]

    Econlogicqa: A question-answering benchmark for evaluating large language models in economic sequential reasoning

    Yinzhu Quan and Zefang Liu. Econlogicqa: A question-answering benchmark for evaluating large language models in economic sequential reasoning. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 2273–2282, 2024

    2024

  54. [62]

    STEER: Assessing the economic rationality of large language models

    Narun Krishnamurthi Raman, Taylor Lundy, Samuel Joseph Amouyal, Yoav Levine, Kevin Leyton-Brown, and Moshe Tennenholtz. STEER: Assessing the economic rationality of large language models. InForty-first International Conference on Machine Learning, 2024. URL https://openreview....

    2024

  55. [63]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    2024 arXiv

  56. [64]

    Glee: A unified framework and benchmark for language-based economic environments.arXiv preprint arXiv:2410.05254, 2024

    Eilam Shapira, Omer Madmon, Itamar Reinman, Samuel Joseph Amouyal, Roi Reichart, and Moshe Tennenholtz. Glee: A unified framework and benchmark for language-based economic environments.arXiv preprint arXiv:2410.05254, 2024

    2024

  57. [65]

    Fastcar: Cache attentive replay for fast auto-regressive video generation on the edge.arXiv preprint arXiv:2505.14709, 2025

    Xuan Shen, Weize Ma, Yufa Zhou, Enhao Tang, Yanyue Xie, Zhengang Li, Yifan Gong, Quanyi Wang, Henghui Ding, Yiwei Wang, et al. Fastcar: Cache attentive replay for fast auto-regressive video generation on the edge.arXiv preprint arXiv:2505.14709, 2025

    2025 arXiv

  58. [66]

    Lazydit: Lazy learning for the acceleration of diffusion transformers

    Xuan Shen, Zhao Song, Yufa Zhou, Bo Chen, Yanyu Li, Yifan Gong, Kai Zhang, Hao Tan, Jason Kuen, Henghui Ding, et al. Lazydit: Lazy learning for the acceleration of diffusion transformers. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 20409–2...

    2025

  59. [67]

    Numerical pruning for efficient autoregressive models

    Xuan Shen, Zhao Song, Yufa Zhou, Bo Chen, Jing Liu, Ruiyi Zhang, Ryan A Rossi, Hao Tan, Tong Yu, Xiang Chen, et al. Numerical pruning for efficient autoregressive models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 20418–20426, 2025

    2025

  60. [68]

    Blumberg, Stephen Marcus McAleer, Yaodong Yang, and Jun Wang

    Oliver Slumbers, David Henry Mguni, Stefano B. Blumberg, Stephen Marcus McAleer, Yaodong Yang, and Jun Wang. A game-theoretic framework for managing risk in multi-agent systems. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Sc...

    2023

  61. [69]

    Crossing the reward bridge: Expanding rl with verifiable rewards across diverse domains

    Yi Su, Dian Yu, Linfeng Song, Juntao Li, Haitao Mi, Zhaopeng Tu, Min Zhang, and Dong Yu. Crossing the reward bridge: Expanding rl with verifiable rewards across diverse domains. arXiv preprint arXiv:2503.23829, 2025

    2025 arXiv

  62. [70]

    Game theory meets large language models: A systematic survey.arXiv preprint arXiv:2502.09053, 2025

    Haoran Sun, Yusen Wu, Yukun Cheng, and Xu Chu. Game theory meets large language models: A systematic survey.arXiv preprint arXiv:2502.09053, 2025. 14

    2025 arXiv

  63. [71]

    Gemma 2: Improving open language models at a practical size.arXiv preprint arXiv:2408.00118, 2024

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, et al. Gemma 2: Improving open language models at a practical size.arXiv preprint arXiv:2408.00118, 2024

    2024 arXiv

  64. [72]

    Qwq-32b: Embracing the power of reinforcement learning, March 2025

    Qwen Team. Qwq-32b: Embracing the power of reinforcement learning, March 2025. URL https://qwenlm.github.io/blog/qwq-32b/

    2025

  65. [73]

    Competing large language models in multi-agent gaming environments

    Jen tse Huang, Eric John Li, Man Ho LAM, Tian Liang, Wenxuan Wang, Youliang Yuan, Wenxiang Jiao, Xing Wang, Zhaopeng Tu, and Michael Lyu. Competing large language models in multi-agent gaming environments. InThe Thirteenth International Conference on Learning Representations, ...

    2025

  66. [74]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    2017

  67. [75]

    Trl: Transformer reinforcement learning.https://github.com/huggingface/trl, 2020

    Leandro von Werra, Younes Belkada, Lewis Tunstall, Edward Beeching, Tristan Thrush, Nathan Lambert, Shengyi Huang, Kashif Rasul, and Quentin Gallouédec. Trl: Transformer reinforcement learning.https://github.com/huggingface/trl, 2020

    2020

  68. [76]

    Drupi: Dataset reduction using privileged information.arXiv preprint arXiv:2410.01611, 2024

    Shaobo Wang, Yantai Yang, Shuaiyu Zhang, Chenghao Sun, Weiya Li, Xuming Hu, and Linfeng Zhang. Drupi: Dataset reduction using privileged information.arXiv preprint arXiv:2410.01611, 2024

    2024

  69. [77]

    Data whisperer: Efficient data selection for task-specific llm fine-tuning via few-shot in-context learning.Annual Meeting of the Association for Computational Linguistics, 2025

    Shaobo Wang, Xiangqi Jin, Ziming Wang, Jize Wang, Jiajun Zhang, Kaixin Li, Zichen Wen, Zhong Li, Conghui He, Xuming Hu, and Linfeng Zhang. Data whisperer: Efficient data selection for task-specific llm fine-tuning via few-shot in-context learning.Annual Meeting of the Associat...

    2025

  70. [78]

    Gnothi seauton: Empowering faithful self-interpretability in black-box transformers.International Conference on Learning Representations, 2025

    Shaobo Wang, Hongxuan Tang, Mingyang Wang, Hongrui Zhang, Xuyang Liu, Weiya Li, Xuming Hu, and Linfeng Zhang. Gnothi seauton: Empowering faithful self-interpretability in black-box transformers.International Conference on Learning Representations, 2025

    2025

  71. [79]

    Not all samples should be utilized equally: Towards understanding and improving dataset distillation

    Shaobo Wang, Yantai Yang, Qilong Wang, Kaixin Li, Linfeng Zhang, and Junchi Yan. Not all samples should be utilized equally: Towards understanding and improving dataset distillation. Synthetic Data for Computer Vision Workshop at CVPR, 2025

    2025

  72. [80]

    Dataset distillation with neural characteristic function: A minmax perspective

    Shaobo Wang, Yicun Yang, Zhiyuan Liu, Chenghao Sun, Xuming Hu, Conghui He, and Linfeng Zhang. Dataset distillation with neural characteristic function: A minmax perspective. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025

    2025

  73. [81]

    Dataset distillation

    Tongzhou Wang, Jun-Yan Zhu, Antonio Torralba, and Alexei A Efros. Dataset distillation. arXiv preprint arXiv:1811.10959, 2018

    2018 arXiv

  74. [82]

    Reinforcement learning for reasoning in large language models with one training example.arXiv preprint arXiv:2504.20571, 2025

    Yiping Wang, Qing Yang, Zhiyuan Zeng, Liliang Ren, Lucas Liu, Baolin Peng, Hao Cheng, Xuehai He, Kuan Wang, Jianfeng Gao, et al. Reinforcement learning for reasoning in large language models with one training example.arXiv preprint arXiv:2504.20571, 2025

    2025 arXiv

  75. [83]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824–24837, 2022

    2022

  76. [84]

    Stop looking for important tokens in multimodal language models: Duplication matters more.arXiv preprint arXiv:2502.11494, 2025

    Zichen Wen, Yifeng Gao, Shaobo Wang, Junyuan Zhang, Qintong Zhang, Weijia Li, Conghui He, and Linfeng Zhang. Stop looking for important tokens in multimodal language models: Duplication matters more.arXiv preprint arXiv:2502.11494, 2025

    2025 arXiv

  77. [85]

    Grounded persuasive language generation for automated marketing.arXiv preprint arXiv:2502.16810, 2025

    Jibang Wu, Chenghao Yang, Simon Mahns, Chaoqi Wang, Hao Zhu, Fei Fang, and Haifeng Xu. Grounded persuasive language generation for automated marketing.arXiv preprint arXiv:2502.16810, 2025

    2025 arXiv

  78. [86]

    Less: Selecting influential data for targeted instruction tuning.arXiv preprint arXiv:2402.04333, 2024

    Mengzhou Xia, Sadhika Malladi, Suchin Gururangan, Sanjeev Arora, and Danqi Chen. Less: Selecting influential data for targeted instruction tuning.arXiv preprint arXiv:2402.04333, 2024. 15

    2024 arXiv

  79. [87]

    Rethinking data selection at scale: Random selection is almost all you need.arXiv preprint arXiv:2410.09335, 2024

    Tingyu Xia, Bowen Yu, Kai Dang, An Yang, Yuan Wu, Yuan Tian, Yi Chang, and Junyang Lin. Rethinking data selection at scale: Random selection is almost all you need.arXiv preprint arXiv:2410.09335, 2024

    2024 arXiv

  80. [88]

    Tradingagents: Multi-agents llm financial trading framework.arXiv preprint arXiv:2412.20138, 2025

    Yijia Xiao, Edward Sun, Di Luo, and Wei Wang. Tradingagents: Multi-agents llm financial trading framework.arXiv preprint arXiv:2412.20138, 2025

    2025 arXiv

  81. [89]

    Logic-rl: Unleashing llm reasoning with rule-based reinforcement learning.arXiv preprint arXiv:2502.14768, 2025

    Tian Xie, Zitian Gao, Qingnan Ren, Haoming Luo, Yuqian Hong, Bryan Dai, Joey Zhou, Kai Qiu, Zhirong Wu, and Chong Luo. Logic-rl: Unleashing llm reasoning with rule-based reinforcement learning.arXiv preprint arXiv:2502.14768, 2025

    2025 arXiv

  82. [90]

    Towards large reasoning models: A survey of reinforced reasoning with large language models.arXiv preprint arXiv:2501.09686, 2025

    Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, et al. Towards large reasoning models: A survey of reinforced reasoning with large language models.arXiv preprint arXiv:2501.09686, 2025

    2025 arXiv

  83. [91]

    Rethinking dataset pruning from a generalization perspective

    Furui Xu, Shaobo Wang, Luo Zhongwei, and Linfeng Zhang. Rethinking dataset pruning from a generalization perspective. InThe Future of Machine Learning Data Practices and Repositories at ICLR 2025, 2025

    2025

  84. [92]

    Karlsson

    Xinrun Xu, Yuxin Wang, Chaoyi Xu, Ziluo Ding, Jiechuan Jiang, Zhiming Ding, and Börje F. Karlsson. A survey on game playing agents and large models: Methods, applications, and challenges, 2024

    2024

  85. [93]

    An Yang, Beichen Zhang, Binyuan Hui, Bofei Gao, Bowen Yu, Chengpeng Li, Dayiheng Liu, Jianhong Tu, Jingren Zhou, Junyang Lin, et al. Qwen2. 5-math technical report: Toward mathematical expert model via self-improvement.arXiv preprint arXiv:2409.12122, 2024

    2024 arXiv

  86. [94]

    Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    2025 arXiv

  87. [95]

    Limo: Less is more for reasoning.arXiv preprint arXiv:2502.03387, 2025

    Yixin Ye, Zhen Huang, Yang Xiao, Ethan Chern, Shijie Xia, and Pengfei Liu. Limo: Less is more for reasoning.arXiv preprint arXiv:2502.03387, 2025

    2025 arXiv

  88. [96]

    Yangyang Yu, Zhiyuan Yao, Haohang Li, Zhiyang Deng, Yuechen Jiang, Yupeng Cao, Zhi Chen, Jordan Suchow, Zhenyu Cui, Rong Liu, et al. Fincon: A synthesized llm multi-agent system with conceptual verbal reinforcement for enhanced financial decision making.Advances in Neural Info...

    2024

  89. [97]

    Synergistic multi-agent framework with trajectory learning for knowledge-intensive tasks

    Shengbin Yue, Siyuan Wang, Wei Chen, Xuanjing Huang, and Zhongyu Wei. Synergistic multi-agent framework with trajectory learning for knowledge-intensive tasks. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 25796–25804, 2025

    2025

  90. [98]

    Does reinforcement learning really incentivize reasoning capacity in llms beyond the base model?arXiv preprint arXiv:2504.13837, 2025

    Yang Yue, Zhiqi Chen, Rui Lu, Andrew Zhao, Zhaokai Wang, Shiji Song, and Gao Huang. Does reinforcement learning really incentivize reasoning capacity in llms beyond the base model?arXiv preprint arXiv:2504.13837, 2025

    2025 arXiv

  91. [99]

    Multi-agent reinforcement learning: A selective overview of theories and algorithms

    Kaiqing Zhang, Zhuoran Yang, and Tamer Ba¸ sar. Multi-agent reinforcement learning: A selective overview of theories and algorithms. InHandbook of Reinforcement Learning and Control, pages 321–384. Springer, 2021

    2021

  92. [100]

    Med-rlvr: Emerging medical reasoning from a 3b base model via reinforcement learning.arXiv preprint arXiv:2502.19655, 2025

    Sheng Zhang, Qianchu Liu, Guanghui Qin, Tristan Naumann, and Hoifung Poon. Med-rlvr: Emerging medical reasoning from a 3b base model via reinforcement learning.arXiv preprint arXiv:2502.19655, 2025

    2025 arXiv

  93. [101]

    LLM as a mastermind: A survey of strategic reasoning with large language models

    Yadong Zhang, Shaoguang Mao, Tao Ge, Xun Wang, Yan Xia, Wenshan Wu, Ting Song, Man Lan, and Furu Wei. LLM as a mastermind: A survey of strategic reasoning with large language models. InFirst Conference on Language Modeling, 2024. URL https: //openreview.net/forum?id=iMqJsQ4evS

    2024

  94. [102]

    Dataset condensation with gradient matching.arXiv preprint arXiv:2006.05929, 2020

    Bo Zhao, Konda Reddy Mopuri, and Hakan Bilen. Dataset condensation with gradient matching.arXiv preprint arXiv:2006.05929, 2020. 16

    2006 arXiv

  95. [103]

    SOTOPIA: Interactive evaluation for social intelligence in language agents

    Xuhui Zhou, Hao Zhu, Leena Mathur, Ruohong Zhang, Haofei Yu, Zhengyang Qi, Louis- Philippe Morency, Yonatan Bisk, Daniel Fried, Graham Neubig, and Maarten Sap. SOTOPIA: Interactive evaluation for social intelligence in language agents. InThe Twelfth International Conference on...

    2024

  96. [104]

    Option 2

    Jie Zhu, Qian Chen, Huaixia Dou, Junhui Li, Lifan Guo, Feng Chen, and Chi Zhang. Dianjin- r1: Evaluating and enhancing financial reasoning in large language models.arXiv preprint arXiv:2504.15716, 2025. A Appendix A.1 Additional Related Work Advancements in Large Language Mode...

    2025 arXiv

  97. [2023]

    URLhttps://proceedings.mlr.press/v202/slumbers23a.html

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.