Pith. sign in

REVIEW 5 major objections 6 minor 37 references

PREMISE: Scalable and Strategic Prompt Optimization for Efficient Mathematical Reasoning in Large Models

T0 review · 5 major / 6 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 A prompt-only method matches math accuracy while cutting reasoning tokens by up to 87.5%.

desk verdict The paper's headline savings claim is contradicted by its own Table 1, Section 4.2, and abstract, so the central result is not established; the concise-CoT phenomenon itself is real but already covered by prior work. read the letter →

arxiv 2506.10716 v1 pith:BVQMEQTS submitted 2025-06-12 cs.CL cs.AIcs.LG

Ye Yu , Yaoning Yu , Haohan Wang This is my paper
classification cs.CLcs.AIcs.LG
keywords promptoptimizationchain-of-thoughtlargereasoningmodelsoverthinkingunderthinkingtokenefficiencymathematicalblack-boxAPI
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

The paper tries to prove that the long chain-of-thought traces produced by large reasoning models are not a necessary price of accuracy. It introduces PREMISE, a prompt-only framework that measures where traces waste tokens and then optimizes a prompt so the model reasons more tersely while staying correct. Across three math benchmarks and two commercial models, the method matches or improves accuracy while shrinking reasoning tokens by as much as 87.5% and cutting dollar cost by 69–82%. If correct, this makes efficient reasoning available to anyone using a black-box API, with no fine-tuning and no multi-sample decoding.

What carries the argument

The central machinery is the pair of trace-level inefficiency metrics $\Xi_O$ and $\Xi_U$, measuring overthinking and underthinking respectively, combined with a multi-objective natural-language gradient search. Textual gradients for answer correctness and token length are scalarized as $\delta = \lambda \delta_{acc} + (1-\lambda)\delta_{len}$ and applied as prompt updates, so the method navigates the Pareto frontier between accuracy and brevity entirely through a single-pass black-box interface.

What would settle it

Re-run PREMISE on MATH-500 with Gemini-2.5-flash and compare the measured accuracy against the paper's own numbers: the prose reports 82%, while Table 1 reports 96%.

Watch

Extended reading notes

Core claim

The paper's central claim is that a static, optimized prompt can compress the hidden chain-of-thought of commercial black-box reasoning models without touching model weights. PREMISE defines overthinking as the fraction of tokens in a correct trace that exceed the shortest correct trace, and underthinking as the fraction of a trace after the point where it irreversibly leaves a recoverable correct path; it uses these diagnostics in a multi-objective textual search that balances correctness against token length. On GSM8K, SVAMP, and MATH-500, the resulting prompt matches or exceeds standard CoT accuracy for Claude and Gemini while reducing reasoning tokens by up to 87.5% and dollar cost by 69–82%. The paper reports that this is not universal: with OpenAI o1 the thinking tokens and cost rise, and its own prose reports an accuracy drop to 82% for Gemini on MATH-500, a number the table contradicts.

Load-bearing premise

The method assumes that a single static prompt, optimized on held-out traces, can make any black-box reasoning model produce hidden reasoning that is shorter yet still correct.

Editorial extensions

If this is right

  • On GSM8K and SVAMP with Claude and Gemini, PREMISE cuts total token footprint by roughly 75–79% with accuracy within one point, and dollar cost by 69–82%.
  • The same prompt-level compression works inside multi-agent systems: for example, Gemini on GSM8K rises from 85% to 90% accuracy while cost falls by 37%.
  • On models that expose a hidden reasoning channel, the prompt can redirect and compress the trace; on OpenAI o1, which exposes only a single completion stream, PREMISE increases thinking tokens and cost, so the paper expects limited leverage until reasoning usage statistics are exposed.
  • The compression ratio is not universally safe: on proof-heavy MATH-500 with Gemini, an overly concise prompt can skip justifications and drop accuracy by about 14 points, so a difficulty-adaptive token budget is needed.
  • Because the method needs only one forward pass per question and no reranking or decoding overhead, it is deployable in latency-sensitive and API-constrained settings.

Reading between the lines

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

  • A natural extension is difficulty-conditioned prompts: estimate each problem's complexity and choose the compression strength $\lambda$ accordingly, which could recover Gemini's MATH-500 accuracy while retaining most token savings.
  • Since the overthinking and underthinking metrics require access to tokenized reasoning traces, the same diagnostic idea could be adapted to fully closed APIs by using final-answer consistency or completion-length distributions as proxy signals.
  • The reported discrepancy for Gemini on MATH-500 (82% in the prose versus 96% in the table) is worth resolving before relying on the method's accuracy guarantees; a simple rerun of that single benchmark would settle which number represents the method.
  • Because the framework is prompt-only, it could be composed with weight-accessible efficiency techniques in settings where fine-tuning is allowed, potentially multiplying the savings.
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

5 major / 6 minor

Summary. The paper proposes PREMISE, a prompt-only framework intended to reduce chain-of-thought reasoning cost in black-box large reasoning models. It defines trace-level overthinking and underthinking metrics, uses them together with a multi-objective natural-language gradient search to optimize a static prompt, and evaluates the resulting prompt on GSM8K, SVAMP, and MATH-500 with Claude 3.7 Sonnet, OpenAI o1, and Gemini 2.5 Flash, as well as in a multi-agent setting. The advertised result is that PREMISE matches or exceeds baseline accuracy while reducing reasoning tokens by up to 87.5% and dollar cost by 69–82%. The manuscript is internally inconsistent on the central numbers: the abstract, Table 1, and Section 4.2 report mutually incompatible Gemini MATH-500 accuracy values, and the claimed universal cost reduction is contradicted by the same table for OpenAI o1. The method section also relies on quantities that are not computable as defined and does not state whether the optimized prompt was selected on a held-out split.

Significance. If the empirical claims were correct, PREMISE would be a practically valuable result: a single-pass, black-box, prompt-only method that cuts LRM inference cost by roughly four-fifths on standard math benchmarks without accuracy loss would be directly usable with commercial APIs. The idea of combining trace-level diagnostics with textual multi-objective search is a reasonable direction, and the paper includes experiments across several models and an agent setting. However, the central claim is not currently supported by the manuscript's own evidence. The numerical contradictions in the headline results and the lack of a clearly described held-out evaluation protocol mean that the reported savings cannot be trusted as stated. I also see no code or detailed prompt-release that would allow independent verification, which further lowers the evidentiary value of the reported numbers.

major comments (5)
  1. [Abstract, Table 1, Section 4.2] The central experimental claim is internally inconsistent. The abstract reports '91%→92% with Gemini' on MATH-500, Table 1 lists the PREMISE accuracy for Gemini-2.5-flash on MATH-500 as 96% (with Normal at 95%), and Section 4.2 states that 'PREMISE attains only 82% accuracy on MATH-500 with Gemini, a 14% drop relative to the normal CoT run.' These three values cannot be reconciled by sampling noise or a typo: 82% is a 14-point deficit, 96% is a 1-point gain, and 92% is yet another value. Because the 'match or exceed baseline accuracy' claim is the load-bearing part of the paper, this contradiction means the primary result is not established as reported.
  2. [Table 1, Section 4.2, Abstract] The claimed cost reduction is contradicted by the paper's own data for OpenAI o1. Table 1 shows that on GSM8K, PREMISE increases thinking tokens from 249 to 1,012 and per-iteration cost from $0.02280 to $0.07061, and on MATH-500 it increases cost from $0.10938 to $0.13457. Section 4.2 explicitly acknowledges this exception, yet the abstract and conclusion state that PREMISE cuts dollar cost by 69–82% without qualification. The claimed savings are model-dependent, and the abstract's universal formulation is not supported by the reported experiments.
  3. [Sections 3.3 and 3.4] The overthinking and underthinking metrics are not operational as defined. L*(q) in Section 3.3 is defined as the minimum length over the set R of all possible reasoning traces, and k*(r,q) in Section 3.4 requires knowing whether a correct continuation exists for an arbitrary prefix in R. For any realistic language model, R is effectively infinite and the existence of a correct continuation is undecidable or at least requires an oracle. The paper gives no approximation, estimator, or computational procedure for these quantities, so the claimed trace-level diagnostics cannot actually be computed from API outputs. This undermines the methodological foundation of the framework, not just its presentation.
  4. [Section 3.6 and Section 4.1] The evaluation protocol does not establish that the reported results are out-of-sample. The multi-objective search in Section 3.6 directly optimizes the two quantities that are later reported as gains: accuracy and token length. Nowhere in Section 4.1 or Section 4.2 is it stated whether the optimized prompt was selected on a held-out portion of each benchmark, whether the numbers in Table 1 come from the same examples used during optimization, how many optimization runs were performed, or what the variance across runs was. Without a held-out evaluation, the reported token reductions and accuracy matches may simply reflect fitting to the test set. This is a load-bearing omission for any prompt-optimization paper.
  5. [Section 5.3 and Table 2] Additional numerical inconsistencies appear in the multi-agent results. Section 5.3 states that 'Gemini's cost on MATH-500 falls by nearly 70%,' but Table 2 shows Gemini on MATH-500 going from $0.146 to $0.085, which is a 42% reduction, not nearly 70%. Section 5.3 also says that on GSM8K with Claude, PREMISE trims 'more than 1.1k reasoning tokens per problem' and the footnote gives a 19% reduction, which is not consistent with the text's cost-focused framing. These errors compound the impression that the empirical tables and prose were not checked against each other.
minor comments (6)
  1. [Section 3.6] There is a typo in 'gradient-base prompt optimization method'; it should be 'gradient-based'.
  2. [Throughout] The dataset name is inconsistently written as 'SV AMP' in the abstract and body and 'SVAMP' in the table captions and references; please standardize.
  3. [Appendix A] The appendix references 'Figure A.1: PREMISE Generated Efficient Reasoning Prompt,' but the figure content is missing from the manuscript. The actual prompt is central to the method and should be printed in full.
  4. [Section 3.1] The notation q(x_i) in the accuracy definition is confusing: q is described as a reasoning schema, but earlier q denotes a question. Please use distinct symbols for questions and prompts.
  5. [Section 5.1] The text says 'Under the overthinking metric defined in Section 3.4,' but the overthinking metric is defined in Section 3.3; the underthinking metric is in Section 3.4.
  6. [Section 5.2] The statement that for OpenAI 'the accuracy of PREMISE is still slightly higher' is contradicted by Table 1, where PREMISE is below Normal on MATH-500 (97% vs. 98%) and below Normal on SVAMP (96% vs. 97%).

Circularity Check

0 steps flagged · score 2.0 of 10

Circularity score 2/10: no load-bearing circular step is exhibited; the central efficiency claim is an empirical benchmark measurement rather than a prediction derived from the paper's own fitted quantities, but self-cited optimization tools, an under-specified evaluation split, and internal accuracy contradictions prevent a fully clean bill of health.

full rationale

The paper's derivation chain is: define overthinking/underthinking trace metrics, cast concise-but-correct reasoning as a Pareto problem over (1-acc, L(r)), optimize a prompt with textual gradients, and report accuracy, token, and cost numbers on GSM8K, SVAMP, and MATH-500. The reported token reductions and accuracies are indeed the quantities the optimizer in Section 3.6 targets, but reporting objective values after optimization is not itself circular; circularity would require showing these values were presented as held-out predictions without a split. The paper never documents a train/validation/test split or a held-out prompt evaluation, so the reader cannot verify that the headline numbers are not just fitted values; this is a missing-support and correctness risk, not a self-definitional reduction I can exhibit from the equations. The self-citations to Revolve (Zhang et al., 2024) and Promptor (Chen et al., 2025) are used as tools rather than as load-bearing correctness arguments; Revolve is an externally published method and the current benchmark experiments are the actual evidence, so this is minor and not circular. The oracular quantities L*(q) and k*(r,q) in Sections 3.2-3.4 are never computed, and Section 5.1's claim that PREMISE is 'significantly closer to the shortest known correct trace' is illustrative rather than measured; this is unsupported but not circular. Separately, the central accuracy claim is internally inconsistent: Section 4.2 states 'PREMISE attains only 82% accuracy on MATH-500 with Gemini, a 14% drop relative to the normal CoT run,' while Table 1 lists 96% for the same condition and the abstract reports 91% to 92%. That contradiction is a correctness flaw, not a circularity, but it means the headline match-or-exceed claim is not established by the paper's own evidence. Net finding: no load-bearing circular step is identifiable in the written derivation; score 2 reflects the minor self-citations and the fragile experimental support without equating those issues to a circular derivation.

Assumptions & free parameters 2 free parameters · 3 assumptions · 2 invented entities

The formal scaffolding of the paper, including the Pareto-frontier optimization and the two trace-efficiency metrics, is never connected to the actual experiments. The only operational free parameter is the prompt template, whose selection procedure is not disclosed; the optimization method is self-cited prior work.

free parameters (2)
  • lambda (scalarization weight) = not reported
    In Section 3.6, the objectives are combined as delta = lambda * delta_acc + (1 - lambda) * delta_len; the paper says lambda can be tuned but never reports the value used in experiments.
  • textual gradient optimizer hyperparameters = not reported
    The number of optimization steps, the initial prompt, and the LLM used to produce textual gradients are not specified, so the prompt optimization procedure is underdetermined.
assumptions (3)
  • domain assumption The most efficient correct trace is the shortest correct trace (Section 3.2).
    Used to define overthinking inefficiency I_O = (L(r) - L*(q)) / L(r); it equates efficiency with token count, ignoring other costs.
  • ad hoc to paper The set R of all possible reasoning traces and the existence of correct continuations for prefixes are accessible enough to compute k*(r,q).
    Sections 3.3-3.4 define metrics using the shortest correct trace and an existential search over continuations; these quantities are not computable for any real model and are never measured in the experiments.
  • domain assumption Natural-language gradients produced by an LLM optimizer are a valid proxy for true gradients of the scalarized objective.
    Section 3.6 relies on textual gradients delta_acc and delta_len from the authors' prior Revolve method (Zhang et al., 2024); no independent validation of this proxy is given.
invented entities (2)
  • Overthinking inefficiency metric I_O
    purpose: To quantify the proportion of unnecessary tokens in a correct reasoning trace relative to the shortest correct trace.
    Defined in Section 3.3 but never computed; relies on unknown shortest trace L*(q).
  • Underthinking inefficiency metric I_U
    purpose: To measure how early an incorrect trace irreversibly leaves a path that could be completed correctly.
    Defined in Section 3.4 but never computed; requires an existential search over all possible continuations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PREMISE: Scalable and Strategic Prompt Optimization for Efficient Mathematical Reasoning in Large Models." pith.science (2026). https://pith.science/paper/BVQMEQTS

@misc{pith2026250610716,
  author       = {Pith},
  title        = {Pith review of: PREMISE: Scalable and Strategic Prompt Optimization for Efficient Mathematical Reasoning in Large Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BVQMEQTS}},
  note         = {Machine review of arXiv:2506.10716}
}
abstract

Large reasoning models (LRMs) such as Claude 3.7 Sonnet and OpenAI o1 achieve strong performance on mathematical benchmarks using lengthy chain-of-thought (CoT) reasoning, but the resulting traces are often unnecessarily verbose. This inflates token usage and cost, limiting deployment in latency-sensitive or API-constrained settings. We introduce PREMISE (PRompt-based Efficient Mathematical Inference with Strategic Evaluation), a prompt-only framework that reduces reasoning overhead without modifying model weights. PREMISE combines trace-level diagnostics with gradient-inspired prompt optimization to minimize redundant computation while preserving answer accuracy. The approach jointly optimizes brevity and correctness through a multi-objective textual search that balances token length and answer validity. Unlike prior work, PREMISE runs in a single-pass black-box interface, so it can be applied directly to commercial LLMs. On GSM8K, SVAMP, and Math500 we match or exceed baseline accuracy ($96\%\rightarrow96\%$ with Claude, $91\%\rightarrow92\%$ with Gemini) while reducing reasoning tokens by up to $87.5\%$ and cutting dollar cost by $69$--$82\%$. These results show that prompt-level optimization is a practical and scalable path to efficient LRM inference without compromising reasoning quality.

Figures

Figures reproduced from arXiv: 2506.10716 by the authors.

Figure 1
Figure 1. Single model comparison on input, thinking, [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Multi-agent systems comparison on input, [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Comparison of PREMISE with single-objective variants that optimise only token count or only accuracy. slips while still saving tokens. 5.3 Multi-Agent System Setting Analysis [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 4 canonical work pages

  1. [1]

    Aytes, Jinheon Baek, and Sung Ju Hwang

    Simon A. Aytes, Jinheon Baek, and Sung Ju Hwang. 2025. https://arxiv.org/abs/2503.05179 Sketch-of-thought: Efficient llm reasoning with adaptive cognitive-inspired sketching . Preprint, arXiv:2503.05179

    arXiv 2025

  2. [2]

    Ke Chen, Yufei Zhou, Xitong Zhang, and Haohan Wang. 2025. https://arxiv.org/abs/2505.13546 Prompt stability matters: Evaluating and optimizing auto-generated prompt in general-purpose systems . Preprint, arXiv:2505.13546

    arXiv 2025

  3. [3]

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde De Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, and 1 others. 2021. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374

    arXiv 2021

  4. [4]

    Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, and 1 others. 2024. Do not think that much for 2+ 3=? on the overthinking of o1-like llms. arXiv preprint arXiv:2412.21187

    arXiv 2024

  5. [5]

    Jeffrey Cheng and Benjamin Van Durme. 2024. Compressed chain of thought: Efficient reasoning through dense representations. arXiv preprint arXiv:2412.13171

    arXiv 2024

  6. [7]

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, and 1 others. 2021 b . Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168

    arXiv 2021

  7. [8]

    Codeforces. 2025. https://codeforces.com/ Codeforces - competitive programming platform . Accessed: 2025-03-18

    work page 2025

  8. [9]

    Yifu Ding, Wentao Jiang, Shunyu Liu, Yongcheng Jing, Jinyang Guo, Yingjie Wang, Jing Zhang, Zengmao Wang, Ziwei Liu, Bo Du, and 1 others. 2025. Dynamic parallel tree search for efficient llm reasoning. arXiv preprint arXiv:2502.16235

    arXiv 2025

Show all 37 references
  1. [10]

    Yichao Fu, Junda Chen, Siqi Zhu, Zheyu Fu, Zhongdongming Dai, Aurick Qiao, and Hao Zhang. 2024. Efficiently serving llm reasoning programs with certaindex. arXiv preprint arXiv:2412.20993

    2024 arXiv

  2. [11]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783

    2024 arXiv

  3. [12]

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948

    2025 arXiv

  4. [13]

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

    2024 arXiv

  5. [14]

    Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason Weston, and Yuandong Tian. 2024. Training large language models to reason in a continuous latent space. arXiv preprint arXiv:2412.06769

    2024 arXiv

  6. [15]

    Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. Measuring mathematical problem solving with the math dataset. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track

    2021

  7. [16]

    Yu Kang, Xianghui Sun, Liangyu Chen, and Wei Zou. 2024. C3ot: Generating shorter chain-of-thought without compromising effectiveness. arXiv preprint arXiv:2412.11664

    2024 arXiv

  8. [17]

    Ayeong Lee, Ethan Che, and Tianyi Peng. 2025. How well do llms compress their own chain-of-thought? a token complexity approach. arXiv preprint arXiv:2503.01141

    2025 arXiv

  9. [18]

    Baohao Liao, Yuhui Xu, Hanze Dong, Junnan Li, Christof Monz, Silvio Savarese, Doyen Sahoo, and Caiming Xiong. 2025. Reward-guided speculative decoding for efficient llm reasoning. arXiv preprint arXiv:2501.19324

    2025 arXiv

  10. [19]

    Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2024. https://openreview.net/forum?id=v8L0pN6EOi Let's verify step by step . In International Conference on Learning Representations (ICLR)

    2024

  11. [20]

    Tengxiao Liu, Qipeng Guo, Xiangkun Hu, Cheng Jiayang, Yue Zhang, Xipeng Qiu, and Zheng Zhang. 2024. Can language models learn to skip steps? arXiv preprint arXiv:2411.01855

    2024 arXiv

  12. [21]

    Tergel Munkhbat, Namgyu Ho, Seohyun Kim, Yongjin Yang, Yujin Kim, and Se-Young Yun. 2025. Self-training elicits concise reasoning in large language models. arXiv preprint arXiv:2502.20122

    2025 arXiv

  13. [22]

    Learning to reason with llms

    OpenAI. Learning to reason with llms. ://openai.com/index/learning-to-reason-with-llms/. Accessed: 15 March 2025

    2025

  14. [23]

    Arkil Patel, Satwik Bhattamishra, and Navin Goyal. 2021. https://arxiv.org/abs/2103.07191 Are nlp models really able to solve simple math word problems? Preprint, arXiv:2103.07191

    2021 arXiv

  15. [24]

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

    2024

  16. [25]

    Zhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu, Yali Du, and Yulan He. 2025. Codi: Compressing chain-of-thought into continuous space via self-distillation. arXiv preprint arXiv:2502.21074

    2025 arXiv

  17. [26]

    DiJia Su, Hanlin Zhu, Yingchen Xu, Jiantao Jiao, Yuandong Tian, and Qinqing Zheng. 2025. Token assorted: Mixing latent and text tokens for improved language model reasoning. arXiv preprint arXiv:2502.03275

    2025 arXiv

  18. [27]

    Hanshi Sun, Momin Haider, Ruiqi Zhang, Huitao Yang, Jiahao Qiu, Ming Yin, Mengdi Wang, Peter Bartlett, and Andrea Zanette. 2024. Fast best-of-n decoding via speculative rejection. arXiv preprint arXiv:2410.20290

    2024 arXiv

  19. [28]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth \'e e Lacroix, Baptiste Rozi \`e re, Naman Goyal, Eric Hambro, Faisal Azhar, and 1 others. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971

    2023 arXiv

  20. [29]

    Yiming Wang, Pei Zhang, Siyuan Huang, Baosong Yang, Zhuosheng Zhang, Fei Huang, and Rui Wang. 2025. Sampling-efficient test-time scaling: Self-estimating the best-of-n sampling in early decoding. arXiv preprint arXiv:2503.01422

    2025

  21. [30]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824--24837

    2022

  22. [31]

    Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, and 1 others. 2025 a . Towards large reasoning models: A survey of reinforced reasoning with large language models. arXiv preprint arXiv:2501.09686

    2025 arXiv

  23. [32]

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

    2025 arXiv

  24. [33]

    Yuchen Yan, Yongliang Shen, Yang Liu, Jin Jiang, Mengdi Zhang, Jian Shao, and Yueting Zhuang. 2025. Inftythink: Breaking the length limits of long-context reasoning in large language models. arXiv preprint arXiv:2503.06692

    2025

  25. [34]

    Wenkai Yang, Shuming Ma, Yankai Lin, and Furu Wei. 2025. Towards thinking-optimal scaling of test-time compute for llm reasoning. arXiv preprint arXiv:2502.18080

    2025

  26. [35]

    Jintian Zhang, Yuqi Zhu, Mengshu Sun, Yujie Luo, Shuofei Qiao, Lun Du, Da Zheng, Huajun Chen, and Ningyu Zhang. 2025. Lightthinker: Thinking step-by-step compression. arXiv preprint arXiv:2502.15589

    2025

  27. [36]

    Peiyan Zhang, Haibo Jin, Leyang Hu, Xinnuo Li, Liying Kang, Man Luo, Yangqiu Song, and Haohan Wang. 2024. https://arxiv.org/abs/2412.03092 Revolve: Optimizing ai systems by tracking response evolution in textual optimization . Preprint, arXiv:2412.03092

    2024 arXiv

  28. [37]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  29. [38]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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