Pith. sign in

REVIEW 6 major objections 6 minor 38 references

MTMT: Consolidating Multiple Thinking Modes to Form a Thought Tree for Strengthening LLM

T0 review · 6 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read By consolidating five thinking modes into a perplexity-gated thought tree, MTMT raises GPT-4o mini's accuracy on GPQA from 38.8% to 44.0% and on TruthfulQA from 55.4% to 58.5%, without external knowledge or per-dataset prompt engineering.

desk verdict Perplexity-guided thought-tree is a plausible combination, but the paper doesn't validate its central control signal, and the reported gains look fragile. read the letter →

arxiv 2412.03987 v1 pith:5VQBC6V3 submitted 2024-12-05 cs.CL cs.AI

Changcheng Li , Xiangyu Wang , Qiuju Chen , Xiren Zhou , Huanhuan Chen This is my paper · ORCID
classification cs.CLcs.AI
keywords MTMTmulti-thinkingmodesthoughttreeperplexity-gatedexpansionSystem2reasoningzero-shotGPQAbenchmark
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

MTMT is a zero-shot method that tries to make a large language model reason the way a careful person would: instead of answering a hard question in one pass, the model is repeatedly prompted to decompose the task, associate related knowledge, compare options, judge importance, and run counterfactual inferences. Each prompted exchange becomes a node in a thought tree, and nodes whose responses show high model uncertainty, measured by perplexity, are expanded further; once the root question's uncertainty drops below a threshold, the accumulated answer is returned. The paper reports that this raises GPT-4o mini accuracy from 38.8% to 44.0% on the graduate-level GPQA benchmark and from 55.4% to 58.5% on TruthfulQA, beating zero-shot Chain of Thought and few-shot baselines. On GSM8K the method barely changes accuracy, which the paper interprets as a sign that the model is already confident on simpler middle-school math. If the result holds, it offers a general, prompt-agnostic way to extract latent knowledge from an LLM for complex reasoning tasks.

What carries the argument

The load-bearing object is the thought tree (the MTMT graph), whose nodes are sub-questions generated by five named thinking modes: decompose, association, compare, importance, and inference. Expansion is gated by the model's perplexity on its own response: a node whose perplexity exceeds a threshold $PPT_i = PPT_0 + \alpha D(i)$ (where $D(i)$ is the node's depth) is treated as confused and receives further thinking-mode prompts. After child nodes contribute relevant information, the parent is regenerated, and a compare-mode prompt selects the better answer; irrelevant nodes are deactivated. The same threshold controls stopping, so the graph's depth, node count, and accuracy-cost balance are governed by $PPT_0$ and $\alpha$.

What would settle it

Take a labeled subset of GPQA, compute the perplexity of each node's sub-answer, and check its correlation with whether that sub-answer is correct; a near-zero or negative correlation would falsify the load-bearing assumption. The complementary experiment holds node count fixed and replaces perplexity-based expansion with random expansion; if accuracy stays the same, the tree's expansion signal is not what produces the gain.

Watch

Extended reading notes

Core claim

The central discovery MTMT claims is that consolidating multiple human-like thinking modes into a single tree structure strengthens an LLM's reasoning more than any single mode on its own. The method formalizes the model's direct answer as System 1 and the tree search as System 2: the graph repeatedly queries the model with prompts drawn from decomposition, association, comparison, importance, and inference, extracts structured information from each response, and decides whether to expand, regenerate, or deactivate nodes based on the response's perplexity. Iterating this loop lets the model revisit its root answer with new information generated from sub-questions. The reported outcome is that this process yields 44.0% on GPQA, 58.5% on TruthfulQA, and 93.9% on GSM8K, with decomposition and association identified by ablation as the most consequential modes.

Load-bearing premise

The load-bearing premise is that the model's uncertainty score on a sub-answer is a reliable signal that the sub-answer is wrong or incomplete, so expanding the most uncertain nodes will improve the final answer; if that correlation is weak, the tree's expansion and stopping rules lose their claimed benefit.

Editorial extensions

If this is right

  • On GPQA, MTMT reaches 44.0% without external knowledge, a 5.2 percentage-point gain over direct GPT-4o mini and higher than the CoT and few-shot baselines tested.
  • On TruthfulQA, MTMT reaches 58.5%, a 3.1-point gain over direct responses, with no per-dataset prompt rewriting.
  • On GSM8K, accuracy is roughly unchanged at 93.9%, consistent with the claim that the method helps most when the base model is not already confident.
  • Ablation shows decomposition is the most important mode (removing it costs 4.7 points on GPQA and 2.1 on TruthfulQA), with association also contributing 4.5 points on GPQA.
  • Higher temperature hurts accuracy, so the paper sets temperature to 0; accuracy falls as the perplexity threshold rises on GPQA, meaning fewer nodes means less correction.

Reading between the lines

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

  • A sharper test of the mechanism would compare perplexity-gated expansion to random expansion at the same node budget; if accuracy does not fall, the tree's structure and regeneration, not the uncertainty signal, would be the cause of the gains.
  • Because $PPT_0$ is set separately for GPQA (1.25) and TruthfulQA (1.45), a natural extension is to evaluate MTMT on a third benchmark using either dataset's tuned threshold, which would reveal whether the method transfers without parameter retuning.
  • The same graph could host a retrieval node type: instead of generating an association from memory, the model would retrieve external facts, turning MTMT into a structured RAG controller while keeping the perplexity-based stopping rule.
  • The explicit perplexity-to-node-count dial also gives a practical way to budget inference cost in agentic settings, trading a few accuracy points for a target number of API calls.
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

6 major / 6 minor

Summary. The paper proposes MTMT, a multi-thinking-mode tree search method for improving zero-shot LLM reasoning. Starting from a root question, the method repeatedly asks the LLM to answer sub-questions generated by several 'thinking modes' (decompose, association, compare, importance, inference), stores the extracted information in a graph, and uses the perplexity of responses to decide when to expand, regenerate, or deactivate nodes. The final root answer is produced after the perplexity condition is met. Experiments with GPT-4o mini on GPQA, TruthfulQA, and GSM8K report accuracy gains over direct answering and CoT baselines (e.g., 38.8% to 44.0% on GPQA), plus parameter sweeps and ablation studies.

Significance. If the claimed effects hold, MTMT would be a useful addition to the toolbox of zero-shot reasoning methods for LLMs, with the distinctive feature of using multiple heterogeneous reasoning prompts and a graph structure. Strengths of the paper include a concrete algorithm description with prompt templates in the appendix, a systematic parameter sweep, and mode-level ablations. However, the reported gains are modest and not supported by statistical tests, and the central confidence mechanism (perplexity-based expansion) is not validated; the current evidence is therefore suggestive rather than conclusive.

major comments (6)
  1. [Section 4.4, Table 1] The sentence 'our method shows a 5.2% and 3.1% improvement on the TruthfulQA and GPQA datasets, respectively' reverses the dataset labels; Table 1 and Section 1 show GPQA 44.0% (from 38.8%, a 5.2-point gain) and TruthfulQA 58.5% (from 55.4%, a 3.1-point gain). This makes the headline quantitative claim ambiguous and needs to be corrected.
  2. [Section 3.3, Eqs. (6)-(7)] The entire node expansion, regeneration, and deactivation policy is driven by the assumption that response perplexity indicates whether a sub-question has been correctly resolved. The paper provides no evidence for this correlation on the datasets used, and it does not compare against a control that spends the same node budget with a random or fixed expansion rule. Without such a control, the reported gains may be due to additional computation rather than to the perplexity signal. I ask the authors to either validate the perplexity-correctness relationship or include a compute-matched baseline.
  3. [Section 3.3, Eq. (6)] The paper never states how the token probabilities P(t_i | t_1...t_{i-1}) are obtained for the GPT-4o mini API. If the API is used in the standard chat-completion mode, per-token log-probabilities are not returned; if a local model or a logprobs parameter is used, that must be described. As written, the central mechanism is not reproducible.
  4. [Section 4.4, Figures 3-5; Section 4.5] The threshold PPT0 and scaling coefficient α are swept on the same test sets used for the headline results, and the ablation experiments use different parameter values on the two datasets (PPT0=1.25, α=0.1 for GPQA; PPT0=1.45, α=0.05 for TruthfulQA). The paper claims a general strategy that works 'without the need to adjust prompts for different datasets,' but it in fact selects control parameters per dataset. This is a form of test-set fitting that likely overstates the gains; a validation split or a single fixed hyperparameter setting should be used.
  5. [Section 4.4, Table 1] No error bars, confidence intervals, or significance tests are reported, and no repeated runs with different seeds are described. Given the stochasticity from random thinking-mode selection and the modest size of the gains (5.2 and 3.1 percentage points), the results could be within sampling noise; the relevant standard errors for binomial accuracies at n=448 and n=817 are roughly 2.3 and 1.7 percentage points.
  6. [Section 3.1, Eq. (5), Section 3.3] The stopping criterion and final-answer selection are not specified. Section 3.1 says that once 'certain criteria' are met the answer Ak becomes final, but Section 3.3 does not state those criteria, nor how the final root answer is selected when multiple candidate answers have been generated at different iterations. This makes the algorithm underspecified.
minor comments (6)
  1. [Section 4.2] The text says TruthfulQA 'contains 654 test instances' but then says 'We use a total of 817 multiple-choice questions from the dataset.' The standard TruthfulQA benchmark has 817 questions; please clarify whether 654 is a subset or a typo.
  2. [Eq. (8)] The formula for AP sums over all data points j, which does not give an average depth as the figure axes and the text ('AP as the depth of the graph') suggest; it should be divided by n or renamed as a total.
  3. [Table 4] The prompt for 'Choose_answer' contains a typo, 'diferences', and inconsistent spacing around 'answer1' and 'answer2'; also 'The diferences' should be 'The differences'.
  4. [References] The reference 'R OpenAI et al. 2023' is an awkward author string; it should be formatted as 'OpenAI'.
  5. [Section 4.5] The statement 'Other thinking modes show similar effects, causing accuracy drops ranging from 2.7% to 0.6%' mixes the GPQA and TruthfulQA tables; on GPQA the drops are 1.4-2.7%, while 0.6% appears only in the TruthfulQA table.
  6. [Section 3.3.1] The paper says a thinking mode is 'randomly chosen' when not assigned, but it does not specify the random seed or distribution, which affects reproducibility even at temperature 0.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: MTMT is an empirical prompting heuristic, not a derivation, and its benchmark claims are not constructed from its own assumptions.

full rationale

MTMT has no formal derivation chain whose outputs are definitionally equal to its inputs. The perplexity-based node expansion (Eqs. 6-7) is an unvalidated heuristic, but the final accuracies are not computed from the threshold by construction; they depend on the base model's actual answers, so the reported gains are not statistically forced by the threshold itself. No load-bearing self-citations appear in the paper: cited prior work (e.g., Yu et al. 2024 for System 1/2 terminology, ToT, CoT) is contextual, and none of these citations is used to rule out alternatives or to supply a conclusion that the experiments themselves must establish. The main validity concerns are not circularity: the per-dataset selection of (PPT0, alpha) for the final reported numbers is a test-set tuning / selection issue that weakens generalization claims but does not make the accuracy measure equivalent to the fitted parameters, and the Section 4.4 sentence attributing a 5.2% gain to TruthfulQA and 3.1% to GPQA contradicts Table 1 (GPQA gains 5.2%, TruthfulQA gains 3.1%), which is a reporting inconsistency, not a circular step. The paper is best read as an empirical prompting study whose central claim rests on an unvalidated perplexity-to-correctness assumption and on test-set configuration choices, but none of these reduce the reported result to its own axioms by construction.

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

The method introduces no new physical or conceptual entities; it is a prompting strategy. The main assumptions are about the reliability of perplexity as a stopping criterion, the adequacy of the thinking-mode set, and the faithfulness of information extraction. These assumptions are not independently verified in the paper.

free parameters (4)
  • Initial perplexity threshold PPT0 = 1.25 (GPQA), 1.45 (TruthfulQA), not stated for GSM8K
    The threshold for node expansion is tuned per dataset after observing the accuracy curves in Figures 3-5. The paper does not justify these values theoretically.
  • Depth scaling coefficient alpha = 0.1 (GPQA), 0.05 (TruthfulQA), not stated for GSM8K
    Controls how the perplexity threshold grows with tree depth; chosen per dataset in the ablation studies and parameter sweeps.
  • Maximum number of thinking nodes = 30
    Set due to API computational limits. This cap likely affects accuracy on difficult questions, as the paper admits that some GPQA tasks may need more nodes.
  • Sampling temperature = 0
    Set to 0 for all experiments after showing that higher temperatures reduce accuracy (Figure 6).
assumptions (4)
  • domain assumption LLM perplexity on a response is a valid proxy for the model's uncertainty about the sub-question
    The entire node expansion mechanism (Equation 6 and 7) relies on perplexity to decide when to keep generating sub-questions. The paper does not validate this assumption against any external measure of uncertainty or correctness.
  • domain assumption The set of thinking modes (decompose, association, compare, importance, inference) is sufficient to elicit latent knowledge
    The method assumes that these five modes cover the useful reasoning strategies and that their combination improves answers. The ablation study only removes one mode at a time in a single configuration, so the claim of complementarity is not strongly supported.
  • domain assumption The information extraction step (converting free-form responses into structured dictionaries) preserves the essential content
    The pipeline extracts information using a Langchain-generated format; the paper assumes this extraction does not lose critical reasoning content. No analysis of extraction errors is provided.
  • domain assumption The base model GPT-4o mini behaves deterministically at temperature 0 and that this setting is appropriate for all tasks
    All final experiments use temperature 0, but the paper does not examine whether sampling variation at other temperatures could change conclusions. The temperature study only covers GPQA.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MTMT: Consolidating Multiple Thinking Modes to Form a Thought Tree for Strengthening LLM." pith.science (2026). https://pith.science/paper/5VQBC6V3

@misc{pith2026241203987,
  author       = {Pith},
  title        = {Pith review of: MTMT: Consolidating Multiple Thinking Modes to Form a Thought Tree for Strengthening LLM},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5VQBC6V3}},
  note         = {Machine review of arXiv:2412.03987}
}
read the original abstract

Large language models (LLMs) have shown limitations in tasks requiring complex logical reasoning and multi-step problem-solving. To address these challenges, researchers have employed carefully designed prompts and flowcharts, simulating human cognitive processes to enhance LLM performance, such as the Chain of Thought approach. In this paper, we introduce MTMT (Multi-thinking Modes Tree), a novel method that interacts with LLMs to construct a thought tree, simulating various advanced cognitive processes, including but not limited to association, counterfactual thinking, task decomposition, and comparison. By breaking down the original complex task into simpler sub-questions, MTMT facilitates easier problem-solving for LLMs, enabling more effective utilization of the latent knowledge within LLMs. We evaluate the performance of MTMT under different parameter configurations, using GPT-4o mini as the base model. Our results demonstrate that integrating multiple modes of thinking significantly enhances the ability of LLMs to handle complex tasks.

Figures

Figures reproduced from arXiv: 2412.03987 by the authors.

Figure 1
Figure 1. The overall structure of MTMT. Starting from the main question, nodes that are "uncertain" about their [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Flowchart of each thinking node. Generate [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The Impact of P P T0 and α on Acc, ANN, and AP in the TruthfulQA Dataset. 1.25 1.35 1.45 PPT0 0.9 0.92 0.94 0.96 0.98 Acc = 0.05 = 0.1 = 0.2 1.25 1.35 1.45 PPT0 1 1.005 1.01 1.015 1.02 1.025 ANN = 0.05 = 0.1 = 0.2 1.25 1.35 1.45 PPT0 1 1.001 1.002 1.003 1.004 1.005 1.006 AP = 0.05 = 0.1 = 0.2 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The Impact of P P T0 and α on Acc, ANN, and AP in the GSM8K Dataset. perature leads to the model’s inability to extract relevant information in the specified format. There￾fore, in other experiments, we consistently set the default temperature to 0. 4.5 Ablation Study …
Figure 5
Figure 5. Figure 5: The Impact of P P T0 and α on Acc, ANN, and AP in the GPQA Dataset. 0 0.15 0.3 0.45 0.6 0.75 0.9 Temperature 0.36 0.38 0.4 0.42 0.44 Acc GPQA [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Relationship between accuracy and LLM tem [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 11 canonical work pages

  1. [1]

    Anthropic. 2024. https://www.anthropic.com/news/claude-3-5-sonnet Claude 3.5 sonnet . Accessed: 2024-08-23

    work page 2024

  2. [2]

    Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2023. https://arxiv.org/abs/2310.11511 Self-rag: Learning to retrieve, generate, and critique through self-reflection . Preprint, arXiv:2310.11511

    arXiv 2023

  3. [3]

    Edward Y Chang. 2023. https://ieeexplore.ieee.org/abstract/document/10099179 Prompting large language models with the socratic method . In 2023 IEEE 13th Annual Computing and Communication Workshop and Conference (CCWC), pages 0351--0360. IEEE

    work page arXiv 2023

  4. [4]

    David Chiang, Peter Cholak, and Anand Pillay. 2023. https://proceedings.mlr.press/v202/chiang23a.html Tighter bounds on the expressivity of transformer encoders . In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pages 5544--5562. PMLR

    2023

  5. [5]

    Nathaniel D Daw, Yael Niv, and Peter Dayan. 2005. https://doi.org/10.1038/nn1560 Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control . Nature neuroscience, 8(12):1704--1711

    doi:10.1038/nn1560 2005

  6. [6]

    Anthony Dickinson and Bernard Balleine. 2002. https://www.researchgate.net/profile/Louis-Matzel/publication/228032407_Learning_Mutants/links/59e0da5f0f7e9b97fbdd52c4/Learning-Mutants.pdf#page=509 The role of learning in the operation of motivational systems . Stevens’ handbook of experimental psychology, 3:497--533

    work page arXiv 2002

  7. [7]

    Jonathan St BT Evans. 2008. https://doi.org/10.1146/annurev.psych.59.103006.093629 Dual-processing accounts of reasoning, judgment, and social cognition . Annu. Rev. Psychol., 59(1):255--278

    arXiv 2008

  8. [8]

    Guhao Feng, Bohang Zhang, Yuntian Gu, Haotian Ye, Di He, and Liwei Wang. 2023. https://proceedings.neurips.cc/paper_files/paper/2023/file/dfc310e81992d2e4cedc09ac47eff13e-Paper-Conference.pdf Towards revealing the mystery behind chain of thought: A theoretical perspective . In Advances in Neural Information Processing Systems, volume 36, pages 70757--7079...

    work page 2023

Show all 38 references
  1. [9]

    Keith Frankish. 2010. https://doi.org/10.1111/j.1747-9991.2010.00330.x Dual-process and dual-system theories of reasoning . Philosophy Compass, 5(10):914--926

    2010

  2. [10]

    Pengbo Hu, Ji Qi, Xingyu Li, Hong Li, Xinqi Wang, Bing Quan, Ruiyu Wang, and Yi Zhou. 2023. https://arxiv.org/abs/2308.09658 Tree-of-mixed-thought: Combining fast and slow thinking for multi-hop visual reasoning . arXiv preprint arXiv:2308.09658

    2023 arXiv

  3. [11]

    Daniel Kahneman. 2011. http://dspace.vnbrims.org:13000/jspui/bitstream/123456789/2224/1/Daniel-Kahneman-Thinking-Fast-and-Slow-.pdf Thinking, fast and slow . macmillan

    2011

  4. [12]

    Daniel Kahneman, Shane Frederick, et al. 2002. https://d1wqtxts1xzle7.cloudfront.net/68120107/Representativeness_revisited_Attribute_s20210715-13197-1cgpp9a.pdf?1626374439=&response-content-disposition=inline Heuristics and biases: The psychology of intuitive judgment, 49(49-81):74

    2002

  5. [13]

    Simon Killcross and Pam Blundell. 2002. http://digital.casalini.it/9789027297846 Associative representations of emotionally significant outcomes . Emotional cognition, page 13

    2002

  6. [14]

    Takeshi Kojima, Shixiang (Shane) Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. https://proceedings.neurips.cc/paper_files/paper/2022/file/8bb0d291acd4acf06ef112099c16f326-Paper-Conference.pdf Large language models are zero-shot reasoners . In Advances in Neural Inf...

    2022

  7. [15]

    Stephanie Lin, Jacob Hilton, and Owain Evans. 2021. https://arxiv.org/abs/2109.07958 Truthfulqa: Measuring how models mimic human falsehoods . CoRR, abs/2109.07958

    2021 arXiv

  8. [16]

    George Loewenstein and Ted O'Donoghue. 2004. https://ssrn.com/abstract=539843 Animal spirits: Affective and deliberative processes in economic behavior . Available at SSRN 539843

    2004

  9. [17]

    Xiao Ma, Swaroop Mishra, Ahmad Beirami, Alex Beutel, and Jilin Chen. 2023. https://arxiv.org/abs/2306.14308 Let's do a thought experiment: Using counterfactuals to improve moral reasoning . arXiv preprint arXiv:2306.14308

    2023 arXiv

  10. [18]

    OpenAI. 2024. https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/ Gpt-4o mini: advancing cost-efficient intelligence . Accessed: 2024-08-3

    2024

  11. [19]

    R OpenAI et al. 2023. https://arxiv.org/abs/2303.08774 Gpt-4 technical report . ArXiv, 2303:08774

    2023 arXiv

  12. [20]

    Arkil Patel, Satwik Bhattamishra, and Navin Goyal. 2021. https://doi.org/10.18653/v1/2021.naacl-main.168 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 Ling...

    2021 doi

  13. [21]

    Ben Prystawski, Michael Li, and Noah Goodman. 2023. https://proceedings.neurips.cc/paper_files/paper/2023/file/e0af79ad53a336b4c4b4f7e2a68eb609-Paper-Conference.pdf Why think step by step? reasoning emerges from the locality of experience . In Advances in Neural Information Pr...

    2023

  14. [22]

    Machel Reid, Nikolay Savinov, Denis Teplyashin, Dmitry Lepikhin, Timothy Lillicrap, Jean-baptiste Alayrac, Radu Soricut, Angeliki Lazaridou, Orhan Firat, Julian Schrittwieser, et al. 2024. https://arxiv.org/abs/2403.05530 Gemini 1.5: Unlocking multimodal understanding across m...

    2024 arXiv

  15. [23]

    David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. 2023. https://arxiv.org/abs/2311.12022 Gpqa: A graduate-level google-proof q&a benchmark . arXiv preprint arXiv:2311.12022

    2023 arXiv

  16. [24]

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. https://proceedings.neurips.cc/paper_files/paper/2023/file/1b44b878bb782e6954cd888628510e90-Paper-Conference.pdf Reflexion: language agents with verbal reinforcement learning . In Advances...

    2023

  17. [25]

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. https://arxiv.org/abs/2203.11171 Self-consistency improves chain of thought reasoning in language models . Preprint, arXiv:2203.11171

    2023 arXiv

  18. [26]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, brian ichter, Fei Xia, Ed Chi, Quoc V Le, and Denny Zhou. 2022. https://proceedings.neurips.cc/paper_files/paper/2022/file/9d5609613524ecf4f15af0f7b31abca4-Paper-Conference.pdf Chain-of-thought prompting elicits reasoning...

    2022

  19. [27]

    Jason Weston and Sainbayar Sukhbaatar. 2023. https://arxiv.org/abs/2311.11829 System 2 attention (is something you might need too) . Preprint, arXiv:2311.11829

    2023 arXiv

  20. [28]

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. 2023 a . https://proceedings.neurips.cc/paper_files/paper/2023/file/271db9922b8d1f4dd7aaef84ed5ac703-Paper-Conference.pdf Tree of thoughts: Deliberate problem solving with large ...

    2023

  21. [29]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023 b . https://par.nsf.gov/biblio/10451467 React: Synergizing reasoning and acting in language models . In International Conference on Learning Representations (ICLR)

    2023

  22. [30]

    Yao Yao, Zuchao Li, and Hai Zhao. 2024. https://arxiv.org/abs/2305.16582 Beyond chain-of-thought, effective graph-of-thought reasoning in language models . Preprint, arXiv:2305.16582

    2024 arXiv

  23. [31]

    Ping Yu, Jing Xu, Jason Weston, and Ilia Kulikov. 2024. https://arxiv.org/abs/2407.06023 Distilling system 2 into system 1 . Preprint, arXiv:2407.06023

    2024 arXiv

  24. [32]

    differentiation

    Mert Yuksekgonul, Federico Bianchi, Joseph Boen, Sheng Liu, Zhi Huang, Carlos Guestrin, and James Zou. 2024. https://arxiv.org/abs/2406.07496 Textgrad: Automatic "differentiation" via text . Preprint, arXiv:2406.07496

    2024 arXiv

  25. [33]

    Yizhou Zhang, Lun Du, Defu Cao, Qiang Fu, and Yan Liu. 2024. https://arxiv.org/abs/2402.05359 Guiding large language models with divide-and-conquer program for discerning problem solving . arXiv preprint arXiv:2402.05359

    2024 arXiv

  26. [34]

    Xufeng Zhao, Mengdi Li, Wenhao Lu, Cornelius Weber, Jae Hee Lee, Kun Chu, and Stefan Wermter. 2024. https://aclanthology.org/2024.lrec-main.543 Enhancing zero-shot chain-of-thought reasoning in large language models through logic . In Proceedings of the 2024 Joint Internationa...

    2024

  27. [35]

    Chi, Quoc V Le, and Denny Zhou

    Huaixiu Steven Zheng, Swaroop Mishra, Xinyun Chen, Heng-Tze Cheng, Ed H. Chi, Quoc V Le, and Denny Zhou. 2024. https://arxiv.org/abs/2310.06117 Take a step back: Evoking reasoning via abstraction in large language models . Preprint, arXiv:2310.06117

    2024 arXiv

  28. [36]

    Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc Le, and Ed Chi. 2023. https://arxiv.org/abs/2205.10625 Least-to-most prompting enables complex reasoning in large language models . Preprint, arXiv...

    2023 arXiv

  29. [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...

  30. [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 11, 2026 · model on record in the stance chip above.