Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

AdapThink: Adaptive Thinking Preferences for Reasoning Language Model

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read By training inside a 2K-token budget, AdapThink gets a small reasoning model to solve more math problems with fewer tokens than four length-control baselines.

desk verdict AdapThink has a plausible idea and decent results, but the missing reward combination makes it unreproducible; fix that and it's a solid incremental contribution. read the letter →

arxiv 2506.18237 v1 pith:UWWP3LBN submitted 2025-06-23 cs.LG cs.AIcs.CL

Xu Wan , Wei Wang , Wenyue Xu , Wotao Yin , Jie Song , Mingyang Sun This is my paper · ORCID
classification cs.LGcs.AIcs.CL
keywords reinforcementlearningchain-of-thoughtefficiencyadaptivereasoninglengthrewarddesigndiversity-awaresamplingmathematicaloverthinkingunderthinking
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

AdapThink is a post-training method for reasoning language models that tries to make the model think for the right amount of time: more reflection on hard problems, less on easy ones, without being told a fixed token budget. The paper's central claim is that a group-relative reward built from model confidence and counts of two kinds of transition words—'pause-validation' words and 'branch-extension' words—can steer a small distilled reasoning model to higher accuracy and lower token use than four length-control baselines. The authors train with only a 2K-token cap and evaluate at 8K tokens, reporting that AdapThink improves first-attempt accuracy (PASS@1) while cutting average response length. If true, this matters because it points to a lightweight, word-level lever for controlling reasoning depth that does not need a stronger external judge or hand-set length limits.

What carries the argument

The carrying mechanism is the group-relative reasoning preference reward (GRPR), Eq. 3, together with the confidence-constrained diversity sampling of Eq. 8. GRPR uses $\varphi$, the fraction of correct answers in a group, to interpolate with a cosine function between two reward regimes: for $\varphi \le \varphi_{\mathrm{low}}$ it emphasizes completion and length control; for $\varphi \ge \varphi_{\mathrm{high}}$ it adds suppression of branch-extension words. The reward components are normalized deviations from the group mean of length, completion, and branch-extension counts, so the signal is relative to what the current model already produces rather than an absolute target. Diversity-aware sampling first oversamples by a factor $K$, then selects groups that maximize an entropy-weighted score over token length, pause-validation words, and branch-extension words, while enforcing minimum counts of correct and incorrect answers.

What would settle it

Take a benchmark where correct solutions naturally require many branch-extension moves, such as combinatorial proof problems with multiple cases. Train two copies of a reasoning model with AdapThink's reward: one using the hand-picked word list and one using a semantic measure of branching (e.g., the number of distinct proof paths attempted). If the word-count version shows lower PASS@1 or loses its token advantage over the semantic version, the word proxy, rather than the adaptive control, is the limiting factor.

Watch

Extended reading notes

Core claim

The paper claims that reasoning efficiency is better controlled by regulating the preference for reflection-related transition words than by penalizing length directly. On a 1.5B distilled reasoning model, AdapThink's group-relative reasoning preference reward compares each response's token count, completion marker, and branch-extension word count with the average of the correct or incorrect group to which the response belongs, then weights those comparisons by the model's group-level confidence $\varphi$. When the model is confident, the reward suppresses branch-extension words; when it is uncertain, the reward mainly encourages completing the answer while keeping length near the group mean. The training group is built by oversampling and then selecting samples to maximize an entropy-based diversity score over length, pause-validation words, and branch-extension words. Under a 2K-token training cap, the paper reports PASS@1 improvements over GRPO, LCPO, TLB, and CosFn on AIME2025, AIME2024, MATH500, and AMC, with fewer tokens on average, and low n-gram repetition rates that suggest the method does not fall into the reward hacking seen with strict length rewards.

Load-bearing premise

The whole reward rests on the assumption that counts of a fixed set of transition words ('wait', 'hold on', 'check', 'verify' for pause-validation; 'alternatively', 'however', 'another', 'instead' for branch-extension) reliably mark wasteful versus useful reasoning, both before and after training.

Editorial extensions

If this is right

  • Training under a short 2K-token cap transfers to 8K-token evaluation, so budget pressure during training can teach efficient habits rather than truncation.
  • Controlling branch-extension words specifically, rather than all reflection words, is what keeps accuracy high while shortening responses.
  • The confidence-interpolated reward means the same rule automatically lets a weaker model think longer and a stronger model think shorter, without a per-question difficulty estimate.
  • Diversity-aware selection with oversampling improves both accuracy and token efficiency, suggesting that group-relative rewards need diverse groups to avoid collapsing to one pattern.
  • Low 40-gram repetition rates relative to LCPO indicate the reward does not encourage degenerate repetition, a known failure mode of hard length penalties.

Reading between the lines

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

  • The word-count proxy could be extended to semantic-level controls: if reflection quality is judged by content overlap rather than surface words, the same confidence-relative design might work for open-ended or non-mathematical reasoning.
  • Because the reward is defined relative to the model's own current output distribution, it may be portable across model scales without retuning the confidence thresholds, though the paper only demonstrates one 1.5B model.
  • A direct test would be to swap the hand-picked word lists for model-generated synonyms; if accuracy degrades, the specific vocabulary matters more than the underlying behavior.
  • The framework could be combined with learned difficulty estimates or separate confidence models to replace the group-accuracy $\varphi$ with a per-sample uncertainty measure.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes AdapThink, an RL post-training framework for reasoning language models that adaptively controls reflection-related transition words instead of imposing fixed length budgets. It introduces a group-relative reasoning preference reward (GRPR) based on model confidence and correctness-group-relative statistics of token length, completion markers, and 'Branch-Extension' words, and a diversity-aware sampling mechanism that oversamples and then entropy-selects a balanced group. Experiments on AIME2025, AIME2024, MATH500, and AMC with DeepSeek-R1-Distill-Qwen-1.5B report higher PASS@1 and lower average token counts than GRPO, TLB, LCPO, and CosFn under an 8K inference budget, together with ablations and an n-gram repetition analysis.

Significance. If the results hold, AdapThink is a useful contribution to the growing literature on efficient reasoning in RL-trained LLMs: it targets reasoning style rather than raw length, introduces diversity-aware group construction, and reports gains over several length-control baselines on public math benchmarks. The paper also includes multiple ablations (reward components, oversampling factor, curriculum training, word-type variants) that go beyond a single main comparison. However, the central empirical claim is not yet reproducible from the manuscript because the reward that is claimed to drive training is never connected to the GRPO objective, and no variance or seed information is given. The work would be strengthened by releasing code and exact training configurations, including the missing reward-combination rule.

major comments (4)
  1. [§4.1, §5, Algorithm 1 (Eqs. 3, 9, 10)] The paper never defines how the GRPR reward r(x,G,θ) from Eq. (3) enters the GRPO update. Algorithm 1 line 15 says 'Update πθ using Equation 9 with GRPR reward (Equation 3) and accuracy reward,' but Eq. (9) is the standard GRPO objective whose advantage A_i in Eq. (10) is computed from the binary accuracy reward r_i = I(y_i = y) only. No equation, appendix, or hyperparameter table specifies whether the GRPR term is added to r_i with a coefficient, replaces the accuracy reward, or is incorporated into the group-relative normalization. Because every PASS@1 and token count in Table 1 is produced by this undefined update, the headline comparison against LCPO/TLB/CosFn is not reproducible from the paper. Please specify the exact combined advantage or reward formalism, including any coefficient for r(x,G,θ).
  2. [§6.1, Table 1] The claim of a '27% average performance improvement' is not supported by the numbers in Table 1. Averaging the relative PASS@1 gains across the four benchmarks gives approximately 31.6% in one simple aggregation, while other reasonable aggregations give around 20.7%; neither equals 27% unless an unreported weighting is used. Please state precisely how the 27% figure is computed, or remove it. In addition, no standard deviations, confidence intervals, or number of independent seeds are reported for any entry in Tables 1–3, so it is impossible to assess whether the observed differences are statistically meaningful rather than checkpoint noise.
  3. [§3, §4.1, §5 (Evaluation Protocol)] The same hand-picked transition words ('wait', 'hold on', 'check', 'verify' for Pause-Validation; 'alternatively', 'however', 'another', 'instead' for Branch-Extension) are used both in the reward (Eq. 3 via λ_b) and in the evaluation metrics (n_p, n_b). This creates a self-referential loop: the model is rewarded for reducing words that are then counted as evidence of efficiency. The paper should provide a concrete test that the chosen word sets remain a valid proxy after training, e.g., by showing that reductions in these words correlate with human-judged reasoning quality or with held-out efficiency metrics that do not depend on the same lexicon, on datasets not seen during training.
  4. [§4.2, Eq. (5), Table 4] Several implementation details needed to reproduce the method are missing or inconsistent. Eq. (5) defines H_* over bins S but does not define how the bins are constructed for length and word counts beyond 'four equal intervals'; the text says α_l = α_b = α_d = 1 while Eq. (6) lists α_l, α_p, α_b; and Table 4 reports batch size 8 with gradient accumulation 8 while the text states a batch size of 32. Please clarify the binning, the exact weights used, and the effective batch size.
minor comments (4)
  1. [Throughout] There are several typos and inconsistencies: 'Methdology' in the Section 4 heading, 'queations' in Section 5, 'Coressponding author' in the footnote, and inconsistent capitalization of 'DeepSeek' (e.g., 'Deepseek-distilled'). A careful proofread is needed.
  2. [Table 1 / §6.1] The rows for FastCuRL-1.5B-24K and DeepScaler-1.5B-24K are marked with a footnote saying they were originally trained with a 24K token limit and are evaluated here with an 8K limit. This is a legitimate reference comparison, but the paper should explicitly note that those models were not trained for the 8K budget and that the comparison may therefore disadvantage them.
  3. [§E.3, Appendix] The example 'AdapThink Example (from AIME 2025)' reports 4,502 tokens and 19 reflection words, while the base model example reports 7,377 tokens and 46 reflection words. These single examples are illustrative, but the text should note that they are anecdotal and not a substitute for the aggregate statistics.
  4. [§7.2, Table 3] The n-gram repetition analysis is a useful reward-hacking check, but the table would be easier to interpret if it reported the standard deviation across test examples or a statistical test comparing AdapThink's 0.7% with GRPO's 0.3%, since both are very small numbers and the qualitative claim about 'consistently low' rates would benefit from uncertainty quantification.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical RL post-training study whose accuracy claims are judged on external benchmarks, not derived from its reward terms.

full rationale

AdapThink is an empirical RL post-training paper; there is no first-principles derivation whose output is hidden in its inputs. The reward in Eq. 3 is a designed objective (group-relative penalties on length, completion, and hand-selected branch-extension words), and the PASS@1 numbers are measured on external benchmarks (AIME2024/2025, MATH500, AMC), not computed from the reward. The selection of transition words from the base model and their subsequent use both as a reward term (λb) and as an evaluation metric (nb) is a self-referential loop: reducing those word counts is partly guaranteed by the training objective. However, this does not make the central accuracy claim circular, because accuracy is independently evaluated and the method outperforms baselines on those external metrics. The paper's Algorithm 1 leaves the combination of GRPR with the GRPO advantage (Eq. 9/10) unspecified, and the '27% average improvement' statistic is not directly reproducible from Table 1, but these are reproducibility/correctness gaps, not circularity. No load-bearing self-citation or imported uniqueness theorem was found; score 0.

Assumptions & free parameters 8 free parameters · 5 assumptions · 2 invented entities

The central claim rests on several hand-set thresholds and modeling choices. The free parameters are the confidence thresholds, sample-count floors, diversity weights, the oversampling factor, the unreported clip bounds, and the ad hoc transition-word lists. The axioms are domain assumptions about what these counts mean and how they transfer across models and benchmarks.

free parameters (8)
  • phi_low = 0.15
    Confidence threshold below which the reward favors output completion over length control. Set by hand in Section 5; no sensitivity analysis.
  • phi_high = 0.5
    Confidence threshold above which branch-extension words are penalized. Set by hand; no sensitivity analysis.
  • Tmin = 3
    Minimum number of correct samples kept in downsampling. Chosen following [26] to preserve correct samples; not ablated.
  • Fmin = 1
    Minimum number of incorrect samples kept in downsampling. Chosen by hand; not ablated.
  • alpha_l, alpha_p, alpha_b = 1, 1, 1
    Weights in the diversity score H_tot (Eq. 6). Set equal by hand; no ablation on these weights.
  • oversampling factor K = 2
    Number of candidate pools; ablation shows K=2 best, but only against K=1 and K=1.75, not a principled selection.
  • rmin, rmax = not reported
    Clip bounds in Eq. 3 are never given in the hyperparameter table, leaving the reward scale unspecified.
  • transition word lists = 4 pause-validation + 4 branch-extension words
    The specific words are chosen from observation of the base model; this is an ad hoc modeling choice, not derived from theory.
assumptions (5)
  • domain assumption Counting transition words ('wait', 'hold on', 'check', 'verify', 'alternatively', 'however', 'another', 'instead') is a valid proxy for reasoning depth and inefficiency.
    Used to build both the GRPR reward (Eq. 3) and the evaluation metrics in Section 5; no external validation that these words correlate with reasoning quality across models.
  • domain assumption Group-relative normalized deviation rewards (Eq. 2) provide useful training signal.
    The paper assumes that being above or below the group mean for length or transition-word counts is informative; GRPO already uses group baselines, so this is an extension but unproven.
  • domain assumption The confidence phi computed from average correctness in a group reflects the model's capability for that question.
    Eq. 1 defines phi, and all adaptive behavior depends on it. If sample correctness is noisy (small group size), the gating may be unstable.
  • domain assumption Entropy-based diversity (Eq. 5) with four equal bins is a meaningful diversity measure for reasoning processes.
    The bin ranges are not specified, and the assumption that uniform distributions over lengths and word counts equal 'diversity' is not justified.
  • domain assumption RL post-training on the curated 5K question dataset transfers to AIME and MATH500 benchmarks.
    Training data includes AIME 1984-2023 and AMC pre-2023, while evaluation includes AIME 2024/2025 and MATH500; transfer is assumed, not demonstrated beyond the results.
invented entities (2)
  • Pause-Validation transition word set
    purpose: Proxy for self-reflection behavior in CoT.
    Four words chosen from base model outputs in Section 3; no external benchmark shows these words are causally tied to reasoning efficiency.
  • Branch-Extension transition word set
    purpose: Proxy for unproductive exploration and underthinking in CoT.
    Four words chosen ad hoc; the reward penalizes these words (Eq. 3) and the evaluation measures them, so the proxy is both target and metric.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AdapThink: Adaptive Thinking Preferences for Reasoning Language Model." pith.science (2026). https://pith.science/paper/UWWP3LBN

@misc{pith2026250618237,
  author       = {Pith},
  title        = {Pith review of: AdapThink: Adaptive Thinking Preferences for Reasoning Language Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UWWP3LBN}},
  note         = {Machine review of arXiv:2506.18237}
}
read the original abstract

Reinforcement Learning (RL)-based post-training has significantly advanced the complex reasoning capabilities of language models, fostering sophisticated self-reflection processes. However, this ``slow thinking'' paradigm presents a critical challenge to reasoning efficiency: models may expend excessive computation on simple questions and shift reasoning prematurely for complex ones. Previous mechanisms typically rely on static length budgets or predefined rules, lacking the adaptability for varying question complexities and models' evolving capabilities. To this end, we propose AdapThink, an adaptive post-training framework designed to induce more efficient thinking while maintaining the performance of reasoning language models. Specifically, AdapThink incorporates two key mechanisms: 1) A group-relative reward function that leverages model confidence and response's characteristic to dynamically adjust the preference of reflection-related transition words without resorting to a fixed length preference. 2) A diversity-aware sampling mechanism that balances the training group's solution accuracy with reasoning diversity via an entropy-guided score. Experiments on several mathematical reasoning datasets with DeepSeek-distilled models demonstrate AdapThink's advantages in enabling adaptive reasoning patterns and mitigating the inefficiencies.

Figures

Figures reproduced from arXiv: 2506.18237 by the authors.

Figure 1
Figure 1. Distribution analysis for DeepSeek-Distill-Qwen 1.5B (top), 7B (bottom) on MATH-500. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The AdapThink framework architecture, featuring confidence-guided sample selection and [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Training comparison among five post-training frameworks on DeepSeek-Distill-Qwen-1.5B [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Training comparison of AdapThink variants. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Frequency distribution of reflection words in base model’s responses. [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Distribution of reflection words across different post-training methods for correct and [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Training dynamics comparison among three AdapThink variants. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Test comparison among three AdapThink variants across checkpoints at 50-step intervals. [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Mitigating Factual Hallucination in Large Reasoning Models via Mixed-Mode Advantage Regularization

    cs.CL 2026-07 conditional novelty 6.0 of 10

    MARGO mitigates thinking-induced hallucination in large reasoning models by using mixed-mode GRPO rollout groups that compare thinking trajectories against same-model non-thinking references.

  2. Think When Needed: Model-Aware Reasoning Routing for LLM-based Ranking

    cs.IR 2026-01 conditional novelty 6.0 of 10

    A per-instance router trained to predict the utility gain of reasoning decides when an LLM should think before ranking, improving NDCG by up to 6.3% while cutting generation tokens by up to 75%.

Reference graph

Works this paper leans on

31 extracted references · 4 canonical work pages · cited by 2 Pith papers

  1. [1]

    L1: Controlling how long a reasoning model thinks with reinforcement learning.arXiv preprint arXiv:2503.04697, 2025

    Pranjal Aggarwal and Sean Welleck. L1: Controlling how long a reasoning model thinks with reinforcement learning.arXiv preprint arXiv:2503.04697, 2025

    arXiv 2025

  2. [2]

    Do not think that much for 2+ 3=? on the overthinking of o1-like llms.arXiv preprint arXiv:2412.21187, 2024

    Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, et al. Do not think that much for 2+ 3=? on the overthinking of o1-like llms.arXiv preprint arXiv:2412.21187, 2024

    arXiv 2024

  3. [3]

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

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    arXiv 2025

  4. [4]

    Token-budget-aware llm reasoning.arXiv preprint arXiv:2412.18547, 2024

    Tingxu Han, Zhenting Wang, Chunrong Fang, Shiyu Zhao, Shiqing Ma, and Zhenyu Chen. Token-budget-aware llm reasoning.arXiv preprint arXiv:2412.18547, 2024

    arXiv 2024

  5. [5]

    The impact of reasoning step length on large language models

    Mingyu Jin, Qinkai Yu, Dong Shu, Haiyan Zhao, Wenyue Hua, Yanda Meng, Yongfeng Zhang, and Mengnan Du. The impact of reasoning step length on large language models. InACL (Findings), 2024

    work page 2024

  6. [6]

    C3ot: Generating shorter chain-of- thought without compromising effectiveness

    Yu Kang, Xianghui Sun, Liangyu Chen, and Wei Zou. C3ot: Generating shorter chain-of- thought without compromising effectiveness. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 24312–24320, 2025

    2025

  7. [7]

    Vineppo: Unlocking rl potential for llm reasoning through refined credit assignment.arXiv preprint arXiv:2410.01679, 2024

    Amirhossein Kazemnejad, Milad Aghajohari, Eva Portelance, Alessandro Sordoni, Siva Reddy, Aaron Courville, and Nicolas Le Roux. Vineppo: Unlocking rl potential for llm reasoning through refined credit assignment.arXiv preprint arXiv:2410.01679, 2024

    arXiv

  8. [8]

    Llm post-training: A deep dive into reasoning large language models.arXiv preprint arXiv:2502.21321, 2025

    Komal Kumar, Tajamul Ashraf, Omkar Thawakar, Rao Muhammad Anwer, Hisham Cholakkal, Mubarak Shah, Ming-Hsuan Yang, Phillip HS Torr, Fahad Shahbaz Khan, and Salman Khan. Llm post-training: A deep dive into reasoning large language models.arXiv preprint arXiv:2502.21321, 2025

    arXiv 2025

Show all 31 references
  1. [9]

    Adaptivestep: Automatically dividing reasoning step through model confidence.arXiv preprint arXiv:2502.13943, 2025

    Yuliang Liu, Junjie Lu, Zhaoling Chen, Chaofeng Qu, Jason Klein Liu, Chonghan Liu, Zefan Cai, Yunhui Xia, Li Zhao, Jiang Bian, et al. Adaptivestep: Automatically dividing reasoning step through model confidence.arXiv preprint arXiv:2502.13943, 2025

    2025 arXiv

  2. [10]

    Tang, Manan Roongta, Colin Cai, Jeffrey Luo, Li Erran Li, Raluca Ada Popa, and Ion Stoica

    Michael Luo, Sijun Tan, Justin Wong, Xiaoxiang Shi, William Y . Tang, Manan Roongta, Colin Cai, Jeffrey Luo, Li Erran Li, Raluca Ada Popa, and Ion Stoica. Deepscaler: Surpassing o1-preview with a 1.5b model by scaling rl, 2025. Notion Blog

    2025

  3. [11]

    Rethinking rl scaling for vision language models: A transparent, from-scratch framework and comprehensive evaluation scheme.arXiv preprint arXiv:2504.02587, 2025

    Yan Ma, Steffi Chern, Xuyang Shen, Yiran Zhong, and Pengfei Liu. Rethinking rl scaling for vision language models: A transparent, from-scratch framework and comprehensive evaluation scheme.arXiv preprint arXiv:2504.02587, 2025

    2025 arXiv

  4. [12]

    Peft: State-of-the-art parameter-efficient fine-tuning methods

    Sourab Mangrulkar, Sylvain Gugger, Lysandre Debut, Younes Belkada, Sayak Paul, and Benjamin Bossan. Peft: State-of-the-art parameter-efficient fine-tuning methods. https: //github.com/huggingface/peft, 2022

    2022

  5. [13]

    Selfcheck: Using llms to zero-shot check their own step-by-step reasoning.arXiv preprint arXiv:2308.00436, 2023

    Ning Miao, Yee Whye Teh, and Tom Rainforth. Selfcheck: Using llms to zero-shot check their own step-by-step reasoning.arXiv preprint arXiv:2308.00436, 2023

    2023 arXiv

  6. [14]

    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

  7. [15]

    Concise thoughts: Impact of output length on llm reasoning and cost.arXiv preprint arXiv:2407.19825, 2024

    Sania Nayab, Giulio Rossolini, Marco Simoni, Andrea Saracino, Giorgio Buttazzo, Nicolamaria Manes, and Fabrizio Giacomelli. Concise thoughts: Impact of output length on llm reasoning and cost.arXiv preprint arXiv:2407.19825, 2024

    2024 arXiv

  8. [16]

    Learning to reason with LLMs

    OpenAI. Learning to reason with LLMs. https://openai.com/index/ learning-to-reason-with-llms, 2024. Accessed: 2024. 10

    2024

  9. [17]

    The benefits of a concise chain of thought on problem- solving in large language models

    Matthew Renze and Erhan Guven. The benefits of a concise chain of thought on problem- solving in large language models. In2024 2nd International Conference on Foundation and Large Language Models (FLLM), pages 476–483. IEEE, 2024

    2024

  10. [18]

    Self-reflection in llm agents: Effects on problem-solving performance.arXiv preprint arXiv:2405.06682, 2024

    Matthew Renze and Erhan Guven. Self-reflection in llm agents: Effects on problem-solving performance.arXiv preprint arXiv:2405.06682, 2024

    2024 arXiv

  11. [19]

    Self-critiquing models for assisting human evaluators.arXiv preprint arXiv:2206.05802, 2022

    William Saunders, Catherine Yeh, Jeff Wu, Steven Bills, Long Ouyang, Jonathan Ward, and Jan Leike. Self-critiquing models for assisting human evaluators.arXiv preprint arXiv:2206.05802, 2022

    2022 arXiv

  12. [20]

    Satori: Reinforcement learning with chain-of-action-thought enhances llm reasoning via autoregressive search.arXiv preprint arXiv:2502.02508, 2025

    Maohao Shen, Guangtao Zeng, Zhenting Qi, Zhang-Wei Hong, Zhenfang Chen, Wei Lu, Gregory Wornell, Subhro Das, David Cox, and Chuang Gan. Satori: Reinforcement learning with chain-of-action-thought enhances llm reasoning via autoregressive search.arXiv preprint arXiv:2502.02508, 2025

    2025 arXiv

  13. [21]

    Dast: Difficulty-adaptive slow-thinking for large reasoning models

    Yi Shen, Jian Zhang, Jieyun Huang, Shuming Shi, Wenjing Zhang, Jiangze Yan, Ning Wang, Kai Wang, and Shiguo Lian. Dast: Difficulty-adaptive slow-thinking for large reasoning models. arXiv preprint arXiv:2503.04472, 2025

    2025

  14. [22]

    Stop overthinking: A survey on efficient reasoning for large language models.arXiv preprint arXiv:2503.16419, 2025

    Yang Sui, Yu-Neng Chuang, Guanchu Wang, Jiamu Zhang, Tianyi Zhang, Jiayi Yuan, Hongyi Liu, Andrew Wen, Shaochen Zhong, Hanjie Chen, et al. Stop overthinking: A survey on efficient reasoning for large language models.arXiv preprint arXiv:2503.16419, 2025

    2025 arXiv

  15. [23]

    Thoughts are all over the place: On the underthinking of o1-like llms.arXiv preprint arXiv:2501.18585, 2025

    Yue Wang, Qiuzhi Liu, Jiahao Xu, Tian Liang, Xingyu Chen, Zhiwei He, Linfeng Song, Dian Yu, Juntao Li, Zhuosheng Zhang, et al. Thoughts are all over the place: On the underthinking of o1-like llms.arXiv preprint arXiv:2501.18585, 2025

    2025 arXiv

  16. [24]

    Large language models are better reasoners with self-verification

    Yixuan Weng, Minjun Zhu, Fei Xia, Bin Li, Shizhu He, Shengping Liu, Bin Sun, Kang Liu, and Jun Zhao. Large language models are better reasoners with self-verification. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 2550–2575, 2023

    2023

  17. [25]

    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

  18. [26]

    A minimalist approach to llm reasoning: from rejection sampling to reinforce.arXiv preprint arXiv:2504.11343, 2025

    Wei Xiong, Jiarui Yao, Yuhui Xu, Bo Pang, Lei Wang, Doyen Sahoo, Junnan Li, Nan Jiang, Tong Zhang, Caiming Xiong, et al. A minimalist approach to llm reasoning: from rejection sampling to reinforce.arXiv preprint arXiv:2504.11343, 2025

    2025 arXiv

  19. [27]

    Chain of draft: Thinking faster by writing less.arXiv preprint arXiv:2502.18600, 2025

    Silei Xu, Wenhao Xie, Lingxiao Zhao, and Pengcheng He. Chain of draft: Thinking faster by writing less.arXiv preprint arXiv:2502.18600, 2025

    2025 arXiv

  20. [28]

    Demystifying long chain-of-thought reasoning in llms.arXiv preprint arXiv:2502.03373, 2025

    Edward Yeo, Yuxuan Tong, Morry Niu, Graham Neubig, and Xiang Yue. Demystifying long chain-of-thought reasoning in llms.arXiv preprint arXiv:2502.03373, 2025

    2025 arXiv

  21. [29]

    Dapo: An open-source llm reinforcement learning system at scale.arXiv preprint arXiv:2503.14476, 2025

    Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Tiantian Fan, Gaohong Liu, Lingjun Liu, Xin Liu, et al. Dapo: An open-source llm reinforcement learning system at scale.arXiv preprint arXiv:2503.14476, 2025

    2025 arXiv

  22. [30]

    Wait”, “check

    Xiaojiang Zhang, Jinghui Wang, Zifei Cheng, Wenhao Zhuang, Zheng Lin, Minglei Zhang, Shaojie Wang, Yinghan Cui, Chao Wang, Junyi Peng, et al. Srpo: A cross-domain imple- mentation of large-scale reinforcement learning on llm.arXiv preprint arXiv:2504.14286, 2025. 11 A Examples...

    2025 arXiv

  23. [31]

    Branch-Extension

    Let me try that... Alternatively, perhaps I can write the equation as:12x 2 −xy−6y 2 = 0... Alternatively, maybe factor it as(ax+by)(cx+dy) = 0... Alternatively, perhaps factor out a 3: 3(4x2 − 1 3 xy−2y 2) = 0 But that doesn’t seem helpful. Alternatively, maybe factor as(4x+a...

Pith tools

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