Pith. sign in

REVIEW 3 major objections 6 minor 17 cited by

Three execution styles organize silent LLM reasoning

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A survey organizing implicit (silent) reasoning in LLMs into three execution paradigms, plus evidence, benchmarks, and challenges.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A broad, well-organized survey of implicit reasoning, but the three-paradigm taxonomy is too fuzzy to serve as the crisp organizing frame it claims to be. the 3 major comments →

arxiv 2509.02350 v1 pith:LRSSYNEF submitted 2025-09-02 cs.CL cs.AI

Implicit Reasoning in Large Language Models: A Comprehensive Survey

Jindong Li , Yali Fu , Li Fan , Jiahong Liu , Yao Shu , Chengwei Qin , Menglin Yang , Irwin King
show 1 more author
Rex Ying
This is my paper
classification cs.CL cs.AI
keywords implicit reasoninglarge language modelslatent reasoningchain-of-thoughtexecution paradigmslatent optimizationsignal-guided controllayer-recurrent execution
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This survey sets out to give the fragmented literature on implicit reasoning in large language models—reasoning that happens internally without emitting visible intermediate text—a single organizing framework. It argues that methods can be classified by how and where internal computation unfolds into three execution paradigms: latent optimization, signal-guided control, and layer-recurrent execution. It also assembles structural, behavioral, and representation-based evidence that such silent reasoning actually occurs, and reviews how the field evaluates it. If the framework holds, researchers gain a shared vocabulary for comparing methods and for seeing where evidence and evaluation are missing. The stakes are practical: implicit reasoning promises cheaper and faster inference than chain-of-thought, but currently lags it in accuracy.

Core claim

The paper's central claim is that the many implicit-reasoning techniques scattered across the literature are not isolated tricks but instances of three execution paradigms defined by where latent computation happens. In latent optimization, the model's internal representations themselves are trained or adjusted—at the level of tokens, reasoning trajectories, or internal states—so that reasoning is compressed into continuous space. In signal-guided control, special tokens or latent signals such as thinking tokens, pause tokens, planning tokens, or memory/reasoning markers allocate extra internal computation or decide between fast and slow inference modes. In layer-recurrent execution, weights

What carries the argument

The key organizing object is the three-way execution-paradigm taxonomy: latent optimization (adjusting internal representations at token, trajectory, or internal-state granularity), signal-guided control (inserting control tokens or latent signals that modulate computation), and layer-recurrent execution (looping shared layers to refine hidden states). The taxonomy is what carries the argument: it converts a list of loosely related methods into a map of where computation happens, and it supplies the survey's structure for comparing methods, evidence, and benchmarks.

Load-bearing premise

The taxonomy's value depends on the assumption that the three execution paradigms are exhaustive and mutually distinct, and that the efficiency advantage of implicit reasoning survives controlled measurement.

What would settle it

A controlled benchmark that runs the same math and commonsense problems through representative methods from each of the three categories, measuring accuracy, latency, tokens, and FLOPs under matched training budgets. If the categories fail to separate—with methods consistently fitting two or three—or if implicit methods lose their speed advantage when verification and retries are counted, the survey's organizing and efficiency claims would be undermined.

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

If this is right

  • Implicit reasoning can reduce decoding cost and latency because no intermediate tokens are emitted, making it attractive for resource-constrained or latency-sensitive applications.
  • The three paradigms are complementary and can be combined, so future systems may mix latent optimization, control signals, and recurrent execution in a single model.
  • Current evidence suggests that trained implicit reasoning can genuinely compute intermediate results, while prompted implicit reasoning often relies on shortcuts rather than robust generalization.
  • Evaluation is underdeveloped: with over seventy datasets used inconsistently, standardized benchmarks and metrics beyond final-answer accuracy are needed to measure reasoning depth and reliability.
  • Future work should develop hybrid explicit/implicit strategies and latent-level supervision signals to close the accuracy gap between implicit and explicit reasoning.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the taxonomy is accepted, it predicts where new methods will land and highlights a missing category: architecture-agnostic approaches that work across model families without task-specific latent heads or loop controllers.
  • The efficiency advantage is the least directly tested claim in the survey; a systematic, controlled comparison of representative methods from each paradigm under equal compute and training budgets would be a natural next step.
  • The evidence section suggests a testable diagnostic: trained implicit reasoning may pass probing tests for intermediate results while prompted implicit reasoning may not, offering a practical way to distinguish genuine latent computation from shortcut behavior.
  • The 'where computation unfolds' axis could extend naturally to multimodal and agentic settings, where silent reasoning may interact with perception, tool use, and memory in ways the current text-focused taxonomy does not yet cover.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper surveys implicit reasoning in large language models, defined as multi-step inference performed internally without emitting explicit intermediate textual steps. It introduces a taxonomy of three 'execution paradigms'—latent optimization, signal-guided control, and layer-recurrent execution—and subdivides them by granularity or control type: token-, trajectory-, and internal-state-level latent optimization; single- and multi-type signal-guided control. The survey also reviews mechanistic, behavioral, and representation-based evidence for implicit reasoning, catalogs evaluation metrics and more than 70 benchmarks, and discusses open challenges. The contribution is primarily organizational: a unified vocabulary for a fragmented literature and a structured map of methods, evidence, and evaluation practices.

Significance. The survey is timely and potentially useful. Its strengths are breadth and systematization: it covers recent 2024–2025 work, provides detailed comparative tables (Tables 2–6 and 7–11), separates method, evidence, and evaluation, and is honest about unresolved issues such as opacity, performance gaps, and lack of standardized evaluation. The proposed execution-paradigm distinction is a plausible organizing idea that could help the community move from per-paper terminology to mechanism-level comparison, and the maintained project page is a useful resource. However, the value of the taxonomy depends on the paradigms being well-defined and the method assignments being recoverable; the current definitions and example classifications are not consistently separable. The efficiency motivation is also stated more categorically than the surveyed evidence supports. With a revised definitional framework and a cleanup of production errors, this could become a valuable reference.

major comments (3)
  1. [§3, esp. §§3.1, 3.1.1, 3.2, 3.2.1, Table 5] The three paradigms are presented as a taxonomy, but the definitions do not yield a recoverable assignment rule. §3.1 defines latent optimization as 'directly adjusting and optimizing internal representations'; §3.1.1 places token-insertion methods (CoCoMix, Latent Token, Token Assorted) in this class. §3.2 defines signal-guided control as 'inserting specialized tokens that modulate computation.' Since inserted tokens are internal representations, the operative mechanisms overlap. Concretely, LatentSeek (Table 5, §3.2.1) is described as 'iteratively optimizes token-wise latent representations via self-rewarding policy gradient at test time,' which satisfies the §3.1 definition, and Heima's thinking token is filed under progressive refinement (§3.1.2(c)) while Herel & Mikolov's thinking token is under signal-guided control (§3.2.1). 'Distinct yet compatible' does not supply a decision cri
  2. [§2.3, Eqs. (5)-(6)] Definition 2 makes implicit reasoning any generation of a hidden trace h1:L (e.g., hidden activations) followed by an answer. Every standard forward pass in a transformer produces layer-wise hidden states before an answer token, so the definition does not delimit the methods surveyed. The intended notion appears narrower: latent thoughts, control tokens, or recurrent states used as extra internal reasoning steps. Please add criteria (e.g., a trace beyond ordinary layer activations, trained/optimized latent variables, or adaptive computation) so the survey's scope is well defined.
  3. [§2.4, Table 1; §§3.3, 3.2.1, 5.1.2, 6] The paper motivates implicit reasoning by efficiency ('Compact, faster, resource-efficient' in Table 1; 'lower generation cost, faster inference' in the abstract), but this is categorical rather than method-dependent. Included methods such as Huginn (§3.3) scale test-time compute, and LatentSeek (§3.2.1) performs iterative test-time optimization; §5.1.2 itself notes that dynamic computation paths make standard efficiency metrics insufficient, and §6 concedes a performance gap versus explicit reasoning. Please qualify the efficiency claims to specific mechanisms and support them with the comparative results from the cited papers, or remove them from Table 1 as inherent properties.
minor comments (6)
  1. [Figure 1] In the explicit-reasoning example, 'Step 4' appears before 'Step 3', and 'markets' should be 'markers'. Fix step ordering and typo.
  2. [Figures 4 and 6] Stray Chinese text ('改正版', '最终版') appears in figure headings; remove or translate.
  3. [Figure 2; §2.4] Typos: 'Distributional Reasoing' in Figure 2; 'need to sample tokens form a finite vocabulary' in §2.4.
  4. [Table 3] CoLaR's base model is listed as LLaMA-3.2-1B-Instruct (Grattafiori et al., 2024); the appropriate source is Meta (2024), consistent with other rows.
  5. [§6] The claim that 'over 70 datasets have been used in isolation' is unsupported; include a count or appendix, or soften to 'many datasets'.
  6. [§2.2–2.3, Eqs. (2), (5)] The notation z1:M ∼ πθ(·|x) and h1:L ∼ πθ(·|x) is informal because πθ defines a distribution over tokens, not hidden states; consider a generic state-generation notation.

Circularity Check

0 steps flagged

No significant circularity; the survey's taxonomy is organizational and its claims rest on external, independently published work.

full rationale

This is a survey paper, so the relevant 'derivation chain' is the construction of the implicit-reasoning taxonomy (Section 3) and the evidence review (Section 4). No quantity is fitted and no prediction is generated from fitted parameters. Definition 2 of implicit reasoning (hidden trace h1:L, final answer a) is a stipulated definition rather than a derived result, and the three execution paradigms are explicitly introduced as an organizing scheme for existing external methods: 'We organize existing methods into three execution paradigms based on how and where internal computation unfolds' (Section 3). The mechanistic and behavioral evidence in Section 4 is cited from prior independent studies (e.g., Din et al. 2024, Wang et al. 2024a, Brinkmann et al. 2024), and the paper even acknowledges open limitations such as the performance gap versus explicit reasoning (Section 6), which shows the survey is not forcing its claims by definition. There are two incidental self-citations in the introduction — Wang et al. 2024b on recommender systems and Liu et al. 2025a on personalized LLMs — but neither is load-bearing: they are cited only as examples of application areas, not as justification for the taxonomy or for any technical claim. A reader could reasonably dispute the mutual exclusivity of the three paradigm definitions (e.g., LatentSeek is placed under signal-guided control although it iteratively optimizes latent representations), but that is a classification-consistency critique, not circularity. The central organizational contribution does not reduce to its own inputs, so the circularity score is minimal.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

As a survey, the paper contributes no fitted parameters or invented entities. Its load-bearing assumptions are the validity of its taxonomy and the efficiency premise, both assumed rather than demonstrated.

axioms (3)
  • ad hoc to paper The three execution paradigms (latent optimization, signal-guided control, layer-recurrent execution) are exhaustive and mutually exclusive categories for implicit reasoning methods.
    Imposed by the authors in Section 3 as the organizing principle; not derived from the literature and called 'complementary' by the authors, implying overlap.
  • domain assumption Implicit reasoning is more efficient than explicit reasoning.
    Stated in the abstract, Section 2.4, and Table 1, supported by cited works (Hao et al. 2024; Zhang et al. 2025a), but not systematically evaluated in the survey; Section 6 notes a performance gap versus explicit reasoning.
  • domain assumption The surveyed papers are representative of the implicit reasoning literature.
    The survey selects papers for the taxonomy; no clear inclusion/exclusion criteria are stated in Section 3, so representativeness is assumed.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Implicit Reasoning in Large Language Models: A Comprehensive Survey." pith.science (2026). https://pith.science/paper/LRSSYNEF

@misc{pith2026250902350,
  author       = {Pith},
  title        = {Pith review of: Implicit Reasoning in Large Language Models: A Comprehensive Survey},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LRSSYNEF}},
  note         = {Machine review of arXiv:2509.02350}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Large Language Models (LLMs) have demonstrated strong generalization across a wide range of tasks. Reasoning with LLMs is central to solving multi-step problems and complex decision-making. To support efficient reasoning, recent studies have shifted attention from explicit chain-of-thought prompting toward implicit reasoning, where reasoning occurs silently via latent structures without emitting intermediate textual steps. Implicit reasoning brings advantages such as lower generation cost, faster inference, and better alignment with internal computation. Although prior surveys have discussed latent representations in the context of reasoning, a dedicated and mechanism-level examination of how reasoning unfolds internally within LLMs remains absent. This survey fills that gap by introducing a taxonomy centered on execution paradigms, shifting the focus from representational forms to computational strategies. We organize existing methods into three execution paradigms based on \textbf{\textit{how and where internal computation unfolds}}: latent optimization, signal-guided control, and layer-recurrent execution. We also review structural, behavioral and representation-based evidence that supports the presence of implicit reasoning in LLMs. We further provide a structured overview of the evaluation metrics and benchmarks used in existing works to assess the effectiveness and reliability of implicit reasoning. We maintain a continuously updated project at: https://github.com/digailab/awesome-llm-implicit-reasoning.

Figures

Figures reproduced from arXiv: 2509.02350 by Chengwei Qin, Irwin King, Jiahong Liu, Jindong Li, Li Fan, Menglin Yang, Rex Ying, Yali Fu, Yao Shu.

Figure 1
Figure 1. Figure 1: Comparison between explicit and implicit reasoning in LLMs. Explicit reasoning shows each step [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Taxonomy of this paper with representative works. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Token-level latent optimization. Illustration of representative paradigms among diverse strategies [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Two representative distillation methods of internal-state-level latent optimization. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Simplified illustration of layer-recurrent execution for implicit reasoning, which usually reuses the [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 17 Pith papers

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

  1. On the Cost and Benefit of Chain of Thought: A Learning-Theoretic Perspective

    cs.LG 2026-05 unverdicted novelty 7.0

    Chain of Thought risk decomposes into oracle-trajectory benefit and trajectory-mismatch cost, with stability determining bounded, linear, or exponential error growth.

  2. CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning

    cs.CL 2026-05 unverdicted novelty 7.0

    CopT reverses CoT by eliciting a draft answer first then using continuous-embedding contrastive verification and on-policy thinking to reflect and correct, yielding up to 23% higher accuracy and 57% fewer tokens witho...

  3. Latent Abstraction for Retrieval-Augmented Generation

    cs.CL 2026-04 unverdicted novelty 7.0

    LAnR unifies retrieval-augmented generation inside a single LLM by deriving dense retrieval vectors from a [PRED] token's hidden states and using entropy to adaptively stop retrieval, outperforming prior RAG on six QA...

  4. S$^2$GR: Stepwise Semantic-Guided Reasoning in Latent Space for Generative Recommendation

    cs.IR 2026-01 unverdicted novelty 7.0

    S²GR adds stepwise thinking tokens with contrastive supervision on codebook clusters to balance computational focus and ground reasoning paths in generative recommendation.

  5. ThinkOmni: A Reasoning-Driven Omni-Modal LLM Framework for Audio Forgery Detection and Localization

    cs.SD 2026-07 conditional novelty 6.0

    ThinkOmni uses an omni-modal LLM with explicit forensic chain-of-thought to detect and localize audio forgery, and reports strong cross-dataset results.

  6. Efficient Chain-of-Modality Reasoning via Progressive Compression for Spoken Language Models

    cs.CL 2026-07 conditional novelty 6.0

    Spoken math models that emit a 40%-compressed reasoning trace between question and answer beat full-reasoning baselines by ~3 accuracy points while using roughly one third of the text tokens.

  7. Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers

    cs.LG 2026-06 unverdicted novelty 6.0

    LOTUS uses a looped padded Transformer with parallel cross-entropy supervision on gold CoT tokens to match explicit CoT performance at 3B parameters while reducing thought-phase latency 2.5x-6.9x.

  8. Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers

    cs.LG 2026-06 accept novelty 6.0

    A looped padded Transformer with parallel gold-CoT cross-entropy supervision matches explicit CoT accuracy at 3B scale and is 2.5–6.9× faster in the thought phase.

  9. LoRi: Low-Rank Distillation for Implicit Reasoning

    cs.CL 2026-06 unverdicted novelty 6.0

    LoRi distills implicit chain-of-thought by matching low-rank structures in hidden states, raising math-reasoning accuracy toward explicit CoT levels on LLaMA and Qwen models.

  10. The Attentional White Bear Effect in Transformer Language Models

    cs.CL 2026-05 unverdicted novelty 6.0

    Prohibited concepts remain recoverable from hidden states, influence attention routing, and shape generations in transformers under instruction-based suppression.

  11. HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering

    cs.AI 2026-04 unverdicted novelty 6.0

    HypEHR is a hyperbolic embedding model for EHR data that uses Lorentzian geometry and hierarchy-aware pretraining to answer clinical questions nearly as well as large language models but with much smaller size.

  12. SeLaR: Selective Latent Reasoning in Large Language Models

    cs.CL 2026-04 unverdicted novelty 6.0

    SeLaR selectively applies latent soft reasoning in LLMs via entropy gating and contrastive regularization, outperforming standard CoT on five benchmarks without training.

  13. Curriculum Learning for Efficient Chain-of-Thought Distillation via Structure-Aware Masking and GRPO

    cs.LG 2026-02 conditional novelty 6.0

    A three-stage curriculum (masked reconstruction, GRPO compression, teacher-guided rewriting) distills long chain-of-thought into a concise 3B student, reporting 76.19% on GSM8K (up from 64.90%) with 167 output tokens ...

  14. MARS: Multi-hop Adaptive Retrieval and SPARQL Generation for KGQA

    cs.CL 2026-07 conditional novelty 5.0

    MARS answers multi-hop knowledge-graph questions by iteratively retrieving ranked triple patterns and letting an LLM decide when to emit a SPARQL query, beating agentic baselines on QALD-10 without fine-tuning.

  15. Latent Visual States for Efficient Multimodal Reasoning

    cs.CV 2026-06 unverdicted novelty 5.0

    EVA generates adaptive Latent_slot tokens as internal visual thoughts, trained end-to-end with text tokens via D-GSPO on the EVA-230K dataset, claiming performance gains and better inference efficiency.

  16. HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering

    cs.AI 2026-04 conditional novelty 5.0

    A 22M-parameter hyperbolic model answers structured EHR questions with accuracy close to LLM-based systems (EHRXQA 89.5%, MIMIC-Instr 76.0%).

  17. The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlook

    cs.AI 2026-04 accept novelty 5.0

    A large survey organizes latent-space work in language-based models by foundation, evolution, four mechanisms, seven abilities, and open challenges.

Reference graph

Works this paper leans on

90 extracted references · 11 canonical work pages · cited by 15 Pith papers · 1 internal anchor

  1. [1]

    https://artofproblemsolving.com/wiki/index.php/ American_Invitational_Mathematics_Examination

    American invitational mathematics examination. https://artofproblemsolving.com/wiki/index.php/ American_Invitational_Mathematics_Examination. Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadallah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, et al. Phi-3 technical report: A highly capable language model local...

    Pith/arXiv arXiv

  2. [6]

    Accessed: 2025-06-25. Stella Biderman, Hailey Schoelkopf, Quentin Gregory Anthony, Herbie Bradley, Kyle O’Brien, Eric Hallahan, Mohammad Aflah Khan, Shivanshu Purohit, USVSN Sai Prashanth, Edward Raff, et al. Pythia: A suite for analyzing large language models across training and scaling. InInternational Conference on Machine Learning, pp. 2397–2430. PMLR,

    2025

  3. [9]

    One billion word benchmark for measuring progress in statistical language modeling.arXiv preprint arXiv:1312.3005,

    Ciprian Chelba, Tomas Mikolov, Mike Schuster, Qi Ge, Thorsten Brants, Phillipp Koehn, and Tony Robin- son. One billion word benchmark for measuring progress in statistical language modeling.arXiv preprint arXiv:1312.3005,

    Pith/arXiv arXiv

  4. [11]

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

    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, 2025a. Wenhu Chen, Ming Yin, Max Ku, Pan Lu, Yixin Wan, Xueguang Ma, Jianyu Xu, Xinyi Wang, and Tony Xia...

    Pith/arXiv arXiv 2023

  5. [13]

    Quac: Question answering in context

    Eunsol Choi, He He, Mohit Iyyer, Mark Yatskar, Wen-tau Yih, Yejin Choi, Percy Liang, and Luke Zettle- moyer. Quac: Question answering in context. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp. 2174–2184,

    2018

  6. [14]

    Think you have solved question answering? try arc, the ai2 reasoning challenge.arXiv preprint arXiv:1803.05457,

    Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge.arXiv preprint arXiv:1803.05457,

    Pith/arXiv arXiv

  7. [15]

    Training verifiers to solve math word problems

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

    Pith/arXiv arXiv

  8. [16]

    Foundational Contributors, Ahmed El-Kishky, Daniel Selsam, Francis Song, Giambattista Parascandolo, Hongyu Ren, Hunter Lightman, Hyung Won, Ilge Akkaya, Ilya Sutskever, Jason Wei, Jonathan Gordon, Karl Cobbe, Kevin Yu, Lukasz Kondraciuk, Max Schwarzer, Mostafa Rohaninejad, Noam Brown, Shengjia Zhao, Trapit Bansal, Vineet Kosaraju, Wenda Zhou Leadership, J...

    Pith/arXiv arXiv

  9. [17]

    Russell Cooper and Andrew A John.Macroeconomics: Theory through applications

    URLhttps://api.semanticscholar.org/CorpusID:274611667. Russell Cooper and Andrew A John.Macroeconomics: Theory through applications. 2012 Book Archive,

    2012

  10. [18]

    Sparse autoencoders find highly interpretable features in language models.arXiv preprint arXiv:2309.08600,

    Hoagy Cunningham, Aidan Ewart, Logan Riggs, Robert Huben, and Lee Sharkey. Sparse autoencoders find highly interpretable features in language models.arXiv preprint arXiv:2309.08600,

    Pith/arXiv arXiv

  11. [19]

    co/deepseek-ai/DeepSeek-V3.1

    URL https://huggingface. co/deepseek-ai/DeepSeek-V3.1. Hybrid thinking/non-thinking inference, improved tool use and agent tasks, long-context extension (128 K), FP8 format. Yuntian Deng, Kiran Prasad, Roland Fernandez, Paul Smolensky, Vishrav Chaudhary, and Stuart Shieber. Implicit chain of thought reasoning via knowledge distillation.arXiv preprint arXi...

    Pith/arXiv arXiv

  12. [20]

    From explicit cot to implicit cot: Learning to internalize cot step by step.arXiv preprint arXiv:2405.14838,

    Yuntian Deng, Yejin Choi, and Stuart Shieber. From explicit cot to implicit cot: Learning to internalize cot step by step.arXiv preprint arXiv:2405.14838,

    Pith/arXiv arXiv

  13. [21]

    Drop: A reading comprehension benchmark requiring discrete reasoning over paragraphs

    Dheeru Dua, Yizhong Wang, Pradeep Dasigi, Gabriel Stanovsky, Sameer Singh, and Matt Gardner. Drop: A reading comprehension benchmark requiring discrete reasoning over paragraphs. InProceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)...

    2019

  14. [22]

    When chain of thought is necessary, language models struggle to evade monitors

    Scott Emmons, Erik Jenner, David K Elson, Rif A Saurous, Senthooran Rajamanoharan, Heng Chen, Irhum Shafkat, and Rohin Shah. When chain of thought is necessary, language models struggle to evade monitors. arXiv preprint arXiv:2507.05246,

    Pith/arXiv arXiv

  15. [23]

    Thinkless: Llm learns when to think, 2025a

    Gongfan Fang, Xinyin Ma, and Xinchao Wang. Thinkless: Llm learns when to think, 2025a. URLhttps: //arxiv.org/abs/2505.13379. Xi Fang, Weijie Xu, Fiona Anting Tan, Ziqing Hu, Jiani Zhang, Yanjun Qi, Srinivasan H Sengamedu, and Christos Faloutsos. Large language models (llms) on tabular data: Prediction, generation, and understanding-a survey.Transactions o...

    Pith/arXiv arXiv

  16. [24]

    The pile: An 800gb dataset of diverse text for language modeling

    Leo Gao, Stella Biderman, Sid Black, Laurence Golding, Travis Hoppe, Charles Foster, Jason Phang, Horace He, Anish Thite, Noa Nabeshima, et al. The pile: An 800gb dataset of diverse text for language modeling. arXiv preprint arXiv:2101.00027,

    Pith/arXiv arXiv

  17. [25]

    Scaling up test-time compute with latent reasoning: A recurrent depth approach.arXiv preprint arXiv:2502.05171,

    Jonas Geiping, Sean McLeish, Neel Jain, John Kirchenbauer, Siddharth Singh, Brian R Bartoldson, Bhavya Kailkhura, Abhinav Bhatele, and Tom Goldstein. Scaling up test-time compute with latent reasoning: A recurrent depth approach.arXiv preprint arXiv:2502.05171,

    Pith/arXiv arXiv

  18. [26]

    Latent preference coding: Aligning large language models via discrete latent codes.arXiv preprint arXiv:2505.04993,

    Zhuocheng Gong, Jian Guan, Wei Wu, Huishuai Zhang, and Dongyan Zhao. Latent preference coding: Aligning large language models via discrete latent codes.arXiv preprint arXiv:2505.04993,

    Pith/arXiv arXiv

  19. [27]

    Continuous chain of thought enables parallel exploration and reasoning.arXiv preprint arXiv:2505.23648,

    Halil Alperen Gozeten, M Emrullah Ildiz, Xuechen Zhang, Hrayr Harutyunyan, Ankit Singh Rawat, and Samet Oymak. Continuous chain of thought enables parallel exploration and reasoning.arXiv preprint arXiv:2505.23648,

    arXiv

  20. [28]

    The llama 3 herd of models

    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 preprint arXiv:2407.21783,

    Pith/arXiv arXiv

  21. [29]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning

    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,

    Pith/arXiv arXiv

  22. [30]

    Beyond chains of thought: Benchmarking latent-space reasoning abilities in large language models.arXiv preprint arXiv:2504.10615,

    Thilo Hagendorff and Sarah Fabi. Beyond chains of thought: Benchmarking latent-space reasoning abilities in large language models.arXiv preprint arXiv:2504.10615,

    Pith/arXiv arXiv

  23. [31]

    Training large language models to reason in a continuous latent space.arXiv preprint arXiv:2412.06769,

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

    Pith/arXiv arXiv

  24. [32]

    Measuring massive multitask language understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Stein- hardt. Measuring massive multitask language understanding. In International Conference on Learning Representations, 2021a. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathemati...

    Pith/arXiv arXiv

  25. [33]

    Reconsidering overthinking: Penalizing internal and external redundancy in cot reasoning

    35 Jialiang Hong, Taihang Zhen, Kai Chen, Jiaheng Liu, Wenpeng Zhu, Jing Huo, Yang Gao, Depeng Wang, Haitao Wan, Xi Yang, et al. Reconsidering overthinking: Penalizing internal and external redundancy in cot reasoning. arXiv preprint arXiv:2508.02178,

    Pith/arXiv arXiv

  26. [34]

    Learning to solve arithmetic word problems with verb categorization

    Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. Learning to solve arithmetic word problems with verb categorization. InProceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pp. 523–533,

    2014

  27. [36]

    Bridging language and items for retrieval and recommendation.arXiv preprint arXiv:2403.03952,

    Yupeng Hou, Jiacheng Li, Zhankui He, An Yan, Xiusi Chen, and Julian McAuley. Bridging language and items for retrieval and recommendation.arXiv preprint arXiv:2403.03952,

    Pith/arXiv arXiv

  28. [37]

    Gpt-4o system card.arXiv preprint arXiv:2410.21276,

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276,

    Pith/arXiv arXiv

  29. [38]

    Mingyu Jin, Weidi Luo, Sitao Cheng, Xinyi Wang, Wenyue Hua, Ruixiang Tang, William Yang Wang, and Yongfeng Zhang

    URLhttps://arxiv.org/abs/2310.06825. Mingyu Jin, Weidi Luo, Sitao Cheng, Xinyi Wang, Wenyue Hua, Ruixiang Tang, William Yang Wang, and Yongfeng Zhang. Disentangling memory and reasoning ability in large language models.arXiv preprint arXiv:2411.13504,

    Pith/arXiv arXiv

  30. [39]

    Aniruddha Kembhavi, Mike Salvato, Eric Kolve, Minjoon Seo, Hannaneh Hajishirzi, and Ali Farhadi

    Accessed: 2025-06-28. Aniruddha Kembhavi, Mike Salvato, Eric Kolve, Minjoon Seo, Hannaneh Hajishirzi, and Ali Farhadi. A diagram is worth a dozen images. In European conference on computer vision, pp. 235–251. Springer,

    2025

  31. [42]

    Scalable language models with posterior inference of latent thought vectors

    Deqian Kong, Minglu Zhao, Dehong Xu, Bo Pang, Shu Wang, Edouardo Honig, Zhangzhang Si, Chuan Li, Jianwen Xie, Sirui Xie, et al. Scalable language models with posterior inference of latent thought vectors. arXiv preprint arXiv:2502.01567,

    Pith/arXiv arXiv

  32. [43]

    Think-to-talk or talk-to-think? when llms come up with an answer in multi-step reasoning

    Keito Kudo, Yoichi Aoki, Tatsuki Kuribayashi, Shusaku Sone, Masaya Taniguchi, Ana Brassard, Keisuke Sakaguchi, and Kentaro Inui. Think-to-talk or talk-to-think? when llms come up with an answer in multi-step reasoning. arXiv preprint arXiv:2412.01113,

    arXiv

  33. [44]

    Race: Large-scale reading com- prehension dataset from examinations

    Guokun Lai, Qizhe Xie, Hanxiao Liu, Yiming Yang, and Eduard Hovy. Race: Large-scale reading com- prehension dataset from examinations. InProceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pp. 785–794,

    work page 2017

  34. [46]

    Large language models in finance: A survey

    Yinheng Li, Shaofei Wang, Han Ding, and Hang Chen. Large language models in finance: A survey. In Proceedings of the fourth ACM international conference on AI in finance, pp. 374–382, 2023a. Yuanzhi Li, Sébastien Bubeck, Ronen Eldan, Allie Del Giorno, Suriya Gunasekar, and Yin Tat Lee. Text- books are all you need ii: phi-1.5 technical report.arXiv prepri...

    Pith/arXiv arXiv

  35. [48]

    Deepseek-v3 technical report.arXiv preprint arXiv:2412.19437, 2024a

    Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. Deepseek-v3 technical report.arXiv preprint arXiv:2412.19437, 2024a. Jiahong Liu, Zexuan Qiu, Zhongyang Li, Quanyu Dai, Jieming Zhu, Minda Hu, Menglin Yang, and Irwin King. A survey of personalized large language models: Prog...

    Pith/arXiv arXiv

  36. [49]

    Can language models learn to skip steps? In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024b

    Tengxiao Liu, Qipeng Guo, Xiangkun Hu, Cheng Jiayang, Yue Zhang, Xipeng Qiu, and Zheng Zhang. Can language models learn to skip steps? In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024b. Tianqiao Liu, Zui Chen, Zitao Liu, Mi Tian, and Weiqi Luo. Expediting and elevating large language model reasoning via hidden chain-of...

    Pith/arXiv arXiv

  37. [50]

    Cot-valve: Length- compressible chain-of-thought tuning.arXiv preprint arXiv:2502.09601,

    Xinyin Ma, Guangnian Wan, Runpeng Yu, Gongfan Fang, and Xinchao Wang. Cot-valve: Length- compressible chain-of-thought tuning.arXiv preprint arXiv:2502.09601,

    Pith/arXiv arXiv

  38. [51]

    Llama 3.2: Revolutionizing edge ai and vision with open, customizable models

    AI Meta. Llama 3.2: Revolutionizing edge ai and vision with open, customizable models. https://ai. meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/ ,

    work page 2024

  39. [52]

    Shen-Yun Miao, Chao-Chun Liang, and Keh-Yih Su

    Ac- cessed: 2024-06-25. Shen-Yun Miao, Chao-Chun Liang, and Keh-Yih Su. A diverse corpus for evaluating and developing english math word problem solvers. InProceedings of the 58th Annual Meeting of the Association for Computa- tional Linguistics, pp. 975–984,

    work page 2024

  40. [53]

    Can a suit of armor conduct electricity? a new dataset for open book question answering

    Todor Mihaylov, Peter Clark, Tushar Khot, and Ashish Sabharwal. Can a suit of armor conduct electricity? a new dataset for open book question answering. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp. 2381–2391,

    2018

  41. [54]

    Beyond words: A latent memory approach to internal reasoning in llms

    José I Orlicki. Beyond words: A latent memory approach to internal reasoning in llms. arXiv preprint arXiv:2502.21030,

    Pith/arXiv arXiv

  42. [55]

    Arkil Patel, Satwik Bhattamishra, and Navin Goyal. Are nlp models really able to solve simple math word problems? In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 2080–2094,

    work page 2021

  43. [56]

    Measuring and narrowing the compositionality gap in language models

    Ofir Press, Muru Zhang, Sewon Min, Ludwig Schmidt, Noah A Smith, and Mike Lewis. Measuring and narrowing the compositionality gap in language models. InFindings of the Association for Computational Linguistics: EMNLP 2023, pp. 5687–5711,

    2023

  44. [57]

    A survey of efficient reasoning for large reasoning models: Language, multi- modality, and beyond.arXiv preprint arXiv:2503.21614,

    Xiaoye Qu, Yafu Li, Zhaochen Su, Weigao Sun, Jianhao Yan, Dongrui Liu, Ganqu Cui, Daizong Liu, Shuxian Liang, Junxian He, et al. A survey of efficient reasoning for large reasoning models: Language, multi- modality, and beyond.arXiv preprint arXiv:2503.21614,

    arXiv

  45. [58]

    Squad: 100,000+ questions for machine comprehension of text

    Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. Squad: 100,000+ questions for machine comprehension of text. InProceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pp. 2383–2392,

    work page 2016

  46. [59]

    Solving general arithmetic word problems

    Subhro Roy and Dan Roth. Solving general arithmetic word problems. InProceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pp. 1743–1752,

    2015

  47. [60]

    Reasoning to learn from latent thoughts

    Yangjun Ruan, Neil Band, Chris J Maddison, and Tatsunori Hashimoto. Reasoning to learn from latent thoughts. arXiv preprint arXiv:2503.18866,

    arXiv

  48. [61]

    Social iqa: Commonsense reasoning about social interactions

    Maarten Sap, Hannah Rashkin, Derek Chen, Ronan Le Bras, and Yejin Choi. Social iqa: Commonsense reasoning about social interactions. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pp. 4463–4473,

    work page 2019

  49. [62]

    Reasoning with latent thoughts: On the power of looped transformers.arXiv preprint arXiv:2502.17416,

    Nikunj Saunshi, Nishanth Dikkala, Zhiyuan Li, Sanjiv Kumar, and Sashank J Reddi. Reasoning with latent thoughts: On the power of looped transformers.arXiv preprint arXiv:2502.17416,

    Pith/arXiv arXiv

  50. [63]

    Distributional reasoning in llms: Parallel reasoning processes in multi-hop reasoning.arXiv preprint arXiv:2406.13858,

    Yuval Shalev, Amir Feder, and Ariel Goldstein. Distributional reasoning in llms: Parallel reasoning processes in multi-hop reasoning.arXiv preprint arXiv:2406.13858,

    Pith/arXiv arXiv

  51. [64]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models

    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,

    Pith/arXiv arXiv

  52. [65]

    Efficient reasoning with hidden thinking

    Xuan Shen, Yizhou Wang, Xiangxi Shi, Yanzhi Wang, Pu Zhao, and Jiuxiang Gu. Efficient reasoning with hidden thinking. arXiv preprint arXiv:2501.19201, 2025a. Zhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu, Yali Du, and Yulan He. Codi: Compressing chain- of-thought into continuous space via self-distillation.arXiv preprint arXiv:2502.21074, 2025b. Aaroh...

    Pith/arXiv arXiv 2023

  53. [66]

    Token assorted: Mixing latent and text tokens for improved language model reasoning.arXiv preprint arXiv:2502.03275,

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

    Pith/arXiv arXiv

  54. [67]

    Stop overthinking: A survey on efficient reasoning for large language models

    40 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,

    Pith/arXiv arXiv

  55. [68]

    A survey of reasoning with foundation models.arXiv preprint arXiv:2312.11562,

    Jiankai Sun, Chuanyang Zheng, Enze Xie, Zhengying Liu, Ruihang Chu, Jianing Qiu, Jiaqi Xu, Mingyu Ding, Hongyang Li, Mengzhe Geng, et al. A survey of reasoning with foundation models.arXiv preprint arXiv:2312.11562,

    Pith/arXiv arXiv

  56. [69]

    Enhancing latent computation in transformers with latent tokens.arXiv preprint arXiv:2505.12629,

    Yuchang Sun, Yanxi Chen, Yaliang Li, and Bolin Ding. Enhancing latent computation in transformers with latent tokens.arXiv preprint arXiv:2505.12629,

    Pith/arXiv arXiv

  57. [70]

    Challenging big-bench tasks and whether chain-of-thought can solve them

    Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc Le, Ed Chi, Denny Zhou, et al. Challenging big-bench tasks and whether chain-of-thought can solve them. InFindings of the Association for Computational Linguistics: ACL 2023, pp. 13003–13051,

    work page 2023

  58. [71]

    Llm pretraining with continuous concepts.arXiv preprint arXiv:2502.08524,

    JihoonTack, JackLanchantin, JaneYu, AndrewCohen, IliaKulikov, JaniceLan, ShiboHao, YuandongTian, Jason Weston, and Xian Li. Llm pretraining with continuous concepts.arXiv preprint arXiv:2502.08524,

    Pith/arXiv arXiv

  59. [72]

    The web as a knowledge-base for answering complex questions

    Alon Talmor and Jonathan Berant. The web as a knowledge-base for answering complex questions. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pp. 641–651,

    work page 2018

  60. [73]

    Commonsenseqa: A question an- swering challenge targeting commonsense knowledge

    Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. Commonsenseqa: A question an- swering challenge targeting commonsense knowledge. InProceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol- ume 1 (Long and Short Papers), pp. 4149–4158,

    work page 2019

  61. [74]

    Think silently, think fast: Dynamic latent compression of llm reasoning chains.arXiv preprint arXiv:2505.16552,

    Wenhui Tan, Jiaze Li, Jianzhong Ju, Zhenbo Luo, Jian Luan, and Ruihua Song. Think silently, think fast: Dynamic latent compression of llm reasoning chains.arXiv preprint arXiv:2505.16552,

    arXiv

  62. [75]

    Think before recommend: Unleashing the latent reasoning power for sequential recommendation.arXiv preprint arXiv:2503.22675,

    Jiakai Tang, Sunhao Dai, Teng Shi, Jun Xu, Xu Chen, Wen Chen, Wu Jian, and Yuning Jiang. Think before recommend: Unleashing the latent reasoning power for sequential recommendation.arXiv preprint arXiv:2503.22675,

    Pith/arXiv arXiv

  63. [76]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023a. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay ...

    Pith/arXiv arXiv

  64. [77]

    Llava-o1: Let vision language models reason step-by-step

    Guowei Xu, Peng Jin, Li Hao, Yibing Song, Lichao Sun, and Li Yuan. Llava-o1: Let vision language models reason step-by-step. arXiv preprint arXiv:2411.10440,

    Pith/arXiv arXiv

  65. [78]

    To cot or to loop? a formal comparison between chain-of-thought and looped transformers

    Kevin Xu and Issei Sato. To cot or to loop? a formal comparison between chain-of-thought and looped transformers. arXiv preprint arXiv:2505.19245,

    arXiv

  66. [79]

    Softcot++: Test-time scaling with soft chain-of-thought reasoning

    Yige Xu, Xu Guo, Zhiwei Zeng, and Chunyan Miao. Softcot++: Test-time scaling with soft chain-of-thought reasoning. arXiv preprint arXiv:2505.11484, 2025a. Yige Xu, Xu Guo, Zhiwei Zeng, and Chunyan Miao. Softcot: Soft chain-of-thought for efficient reasoning with llms. arXiv preprint arXiv:2502.12134, 2025b. Yibo Yan, Shen Wang, Jiahao Huo, Jingheng Ye, Zh...

    Pith/arXiv arXiv

  67. [80]

    Qwen2 technical report, 2024a

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jianxin Yang, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng He, Junyang Lin, Kai Dang, Keming Lu, Keqin Chen, Kexin Yang, Mei...

    Pith/arXiv arXiv 2018

  68. [81]

    Internal chain-of-thought: Empirical evidence for layer-wise subtask scheduling in llms.arXiv preprint arXiv:2505.14530, 2025b

    Zhipeng Yang, Junzhuo Li, Siyu Xia, and Xuming Hu. Internal chain-of-thought: Empirical evidence for layer-wise subtask scheduling in llms.arXiv preprint arXiv:2505.14530, 2025b. Jiaran Ye, Zijun Yao, Zhidian Huang, Liangming Pan, Jinxin Liu, Yushi Bai, Amy Xin, Liu We- ichuan, Xiaoyin Che, Lei Hou, et al. How does transformer learn implicit reasoning? ar...

    arXiv 2025

  69. [82]

    Zihao Yi, Jiarui Ouyang, Yuwen Liu, Tianhao Liao, Zhe Xu, and Ying Shen

    https://business.yelp.com/data/resources/ open-dataset/. Zihao Yi, Jiarui Ouyang, Yuwen Liu, Tianhao Liao, Zhe Xu, and Ying Shen. A survey on recent advances in llm-based multi-turn dialogue systems.arXiv preprint arXiv:2402.18013,

    Pith/arXiv arXiv

  70. [83]

    Qifan Yu, Zhenyu He, Sijie Li, Xun Zhou, Jun Zhang, Jingjing Xu, and Di He

    In The First Workshop on System-2 Reasoning at Scale, NeurIPS’24, 2024b. Qifan Yu, Zhenyu He, Sijie Li, Xun Zhou, Jun Zhang, Jingjing Xu, and Di He. Enhancing auto-regressive chain-of-thought through loop-aligned reasoning.arXiv preprint arXiv:2502.08482, 2025a. Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fa...

    Pith/arXiv arXiv

  71. [84]

    Don’t overthink it: A survey of efficient r1-style large reasoning models

    43 Linan Yue, Yichao Du, Yizhi Wang, Weibo Gao, Fangzhou Yao, Li Wang, Ye Liu, Ziyu Xu, Qi Liu, Shimin Di, et al. Don’t overthink it: A survey of efficient r1-style large reasoning models. arXiv preprint arXiv:2508.02120, 2025a. Zhenrui Yue, Bowen Jin, Huimin Zeng, Honglei Zhuang, Zhen Qin, Jinsung Yoon, Lanyu Shang, Jiawei Han, and Dong Wang. Hybrid late...

    Pith/arXiv arXiv

  72. [85]

    Pretraining language models to ponder in continuous space

    Boyi Zeng, Shixiang Song, Siyuan Huang, Yixuan Wang, He Li, Ziwei He, Xinbing Wang, Zhiyu Li, and Zhouhan Lin. Pretraining language models to ponder in continuous space. arXiv preprint arXiv:2505.20674,

    arXiv

  73. [86]

    Lightthinker: Thinking step-by-step compression.arXiv preprint arXiv:2502.15589, 2025a

    Jintian Zhang, Yuqi Zhu, Mengshu Sun, Yujie Luo, Shuofei Qiao, Lun Du, Da Zheng, Huajun Chen, and Ningyu Zhang. Lightthinker: Thinking step-by-step compression.arXiv preprint arXiv:2502.15589, 2025a. Peiyuan Zhang, Guangtao Zeng, Tianduo Wang, and Wei Lu. Tinyllama: An open-source small language model. arXiv preprint arXiv:2401.02385,

    arXiv

  74. [87]

    A survey on test-time scaling in large language models: What, how, where, and how well?arXiv preprint arXiv:2503.24235, 2025b

    Qiyuan Zhang, Fuyuan Lyu, Zexu Sun, Lei Wang, Weixu Zhang, Wenyue Hua, Haolun Wu, Zhihan Guo, Yufei Wang, Niklas Muennighoff, et al. A survey on test-time scaling in large language models: What, how, where, and how well?arXiv preprint arXiv:2503.24235, 2025b. Xiang Zhang, Junbo Zhao, and Yann LeCun. Character-level convolutional networks for text classifi...

    Pith/arXiv arXiv

  75. [88]

    Soft thinking: Unlocking the reasoning potential of llms in continuous concept space

    Zhen Zhang, Xuehai He, Weixiang Yan, Ao Shen, Chenyang Zhao, Shuohang Wang, Yelong Shen, and Xin Eric Wang. Soft thinking: Unlocking the reasoning potential of llms in continuous concept space. arXiv preprint arXiv:2505.15778, 2025c. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric...

    Pith/arXiv arXiv

  76. [89]

    Re- inforced mllm: A survey on rl-based reasoning in multimodal large language models

    Guanghao Zhou, Panjia Qiu, Cen Chen, Jie Wang, Zheming Yang, Jian Xu, and Minghui Qiu. Re- inforced mllm: A survey on rl-based reasoning in multimodal large language models. arXiv preprint arXiv:2504.21277,

    Pith/arXiv arXiv

  77. [90]

    Reasoning by superposition: A theoretical perspective on chain of continuous thought.arXiv preprint arXiv:2505.12514, 2025a

    Hanlin Zhu, Shibo Hao, Zhiting Hu, Jiantao Jiao, Stuart Russell, and Yuandong Tian. Reasoning by superposition: A theoretical perspective on chain of continuous thought.arXiv preprint arXiv:2505.12514, 2025a. Rui-Jie Zhu, Tianhao Peng, Tianhao Cheng, Xingwei Qu, Jinfa Huang, Dawei Zhu, Hao Wang, Kaiwen Xue, Xuanliang Zhang, Yong Shan, et al. A survey on l...

    arXiv

  78. [2013]

    Language models are hidden reasoners: Unlocking latent reasoning capabilities via self-rewarding.arXiv preprint arXiv:2411.04282, 2024a

    Haolin Chen, Yihao Feng, Zuxin Liu, Weiran Yao, Akshara Prabhakar, Shelby Heinecke, Ricky Ho, Phil Mui, Silvio Savarese, Caiming Xiong, et al. Language models are hidden reasoners: Unlocking latent reasoning capabilities via self-rewarding.arXiv preprint arXiv:2411.04282, 2024a. Lin Chen, Jinsong Li, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Zehui Chen, Haodon...

    Pith/arXiv arXiv

  79. [2014]

    Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models

    Yifan Hou, Jiaoda Li, Yu Fei, Alessandro Stolfo, Wangchunshu Zhou, Guangtao Zeng, Antoine Bosselut, and Mrinmaya Sachan. Towards a mechanistic interpretation of multi-step reasoning capabilities of language models. arXiv preprint arXiv:2310.14491,

    work page internal anchor Pith review Pith/arXiv arXiv

  80. [2015]

    Mawps: A math word problem repository

    36 Rik Koncel-Kedziorski, Subhro Roy, Aida Amini, Nate Kushman, and Hannaneh Hajishirzi. Mawps: A math word problem repository. InProceedings of the 2016 conference of the north american chapter of the association for computational linguistics: human language technologies, pp. 1152–1157,

    work page 2016

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.