Pith. sign in

REVIEW 4 major objections 3 minor 2 cited by

A Comparative Study of Neurosymbolic AI Approaches to Interpretable Logical Reasoning

T0 review · 4 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The hybrid approach to neurosymbolic AI is more promising for general logical reasoning than the integrative approach, because it produces more interpretable reasoning chains and retains the advantages of LLMs.

desk verdict The abstract asks a good question about neurosymbolic architectures, but the full text is unreadable and the conclusion rests on two self-selected case studies. read the letter →

arxiv 2508.03366 v1 pith:PX7JLNVP submitted 2025-08-05 cs.AI cs.CLcs.LGcs.SC

Michael K. Chen This is my paper
classification cs.AIcs.CLcs.LGcs.SC
keywords neurosymbolicAIlogicalreasoninglargelanguagemodelshybridarchitectureintegrativeinterpretabilitysymbolicsolverdomain-agnostic
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 claims that for domain-agnostic deductive reasoning, the hybrid neurosymbolic approach—where a neural network translates the problem and a separate symbolic solver does the deduction—is more promising than the integrative approach, where symbolic reasoning is embedded inside the neural network. This claim is based on comparing two representative systems: LNN (integrative) and LLM-SS (hybrid). The paper argues that the hybrid approach wins on two counts: its chain of reasoning is more interpretable, and it keeps the benefits of modern LLMs. If the paper is right, future work on general logical reasoning should build on modular, model-agnostic frameworks like the one it proposes.

What carries the argument

The central object is the architecture split between integrative and hybrid neurosymbolic systems. In LLM-SS, a large language model handles natural-language understanding and generation, a separate symbolic solver performs deductive inference, and the two are connected through a translation step; in LNN, the network itself realizes logical inference as differentiable operations, so symbolic reasoning is contained within the neural network. The comparison between these two systems is the machinery that carries the argument: LNN stands for one family, LLM-SS stands for the other, and the contrast in interpretability and retained LLM capability is what supports the verdict in favour of hybrid design.

What would settle it

Run LNN and LLM-SS, together with several other integrative and hybrid systems, on a common suite of domain-agnostic deductive reasoning benchmarks, measuring both accuracy and the interpretability of the produced reasoning chain. If an integrative model matches or exceeds LLM-SS on accuracy while producing equally interpretable reasoning, the paper's preference for the hybrid approach would fail; absent that, the claim stands as argued.

Watch

Extended reading notes

Core claim

The paper's central claim is that the hybrid approach to neurosymbolic AI is more promising for developing general logical reasoning than the integrative approach. In the hybrid design, the LLM acts as an interface that converts a problem into symbolic form, an external solver performs the deterministic reasoning, and the result is translated back; in the integrative design, logic is folded directly into the neural network's computations. Using the Logic Neural Network (LNN) as the integrative representative and the LLM-Symbolic Solver (LLM-SS) as the hybrid representative, the paper demonstrates two advantages for hybrid systems: the reasoning chain is more interpretable, and the system retains the capabilities and advantages of existing LLMs. It then proposes a generalizable framework based on LLM-SS that is modular, model-agnostic, and requires little to no human input.

Load-bearing premise

The whole comparison rests on LNN and LLM-SS being representative, best-in-class examples of the integrative and hybrid approaches respectively, so a fair reader must accept that these two systems stand for their families before accepting the general conclusion.

Editorial extensions

If this is right

  • A system builder choosing between the two families should prefer an architecture in which the symbolic solver is a separate component from the neural network.
  • The proposed modular framework gives a concrete starting point for building hybrid reasoners without retraining the LLM.
  • Interpretability of a neurosymbolic reasoner can be evaluated by looking at the explicit reasoning chain produced by the solver, rather than by inspecting network internals.
  • Because LLM-SS retains the LLM's capabilities, hybrid systems can be upgraded by swapping in better LLMs or better solvers without changing the overall architecture.

Reading between the lines

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

  • A testable extension of the paper's reasoning is that replacing the LLM in LLM-SS with a stronger model should improve end-to-end accuracy without changing the solver, because the hybrid architecture isolates the translation bottleneck.
  • The same modular design could be extended to non-deductive reasoning by swapping the solver for one that handles abduction or induction, which the paper does not explore.
  • If interpretability is defined as an explicit, inspectable reasoning chain, then the hybrid approach may be judged more interpretable by construction; the paper's comparison would then be partly a definitional consequence rather than a purely empirical discovery.
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

4 major / 3 minor

Summary. The paper proposes a taxonomy of neurosymbolic approaches to logical reasoning, distinguishing integrative models, in which symbolic reasoning is contained within the neural network, from hybrid models, in which a separate external symbolic solver performs the reasoning. It introduces two case-study systems, Logic Neural Network (LNN) for the integrative approach and LLM-Symbolic Solver (LLM-SS) for the hybrid approach, and claims that the hybrid approach is more promising for developing general logical reasoning because (i) its reasoning chain is more interpretable and (ii) it retains the capabilities and advantages of existing LLMs. The paper then proposes a modular, model-agnostic framework based on LLM-SS for future work.

Significance. If the central claim were established, the paper would offer a useful design guideline for neurosymbolic reasoning systems and a concrete framework for building them. The taxonomic distinction between integrative and hybrid approaches is clear and potentially valuable, and the proposed modular framework is a constructive contribution. However, the submitted version does not provide verifiable quantitative evidence: the abstract reports no numbers, the readable portions of the text contain no benchmark results, significance tests, or interpretability measurements, and the tables are largely unreadable in the supplied version. The inference from two self-introduced systems to two entire model families is not supported. The paper does not appear to include reproducible code, data, or machine-checked proofs, so the comparative claim cannot be checked from the submission.

major comments (4)
  1. [Abstract] The central claim that 'our analysis demonstrates that the hybrid approach is more promising' is not supported by any quantitative evidence in the abstract or in the readable portions of the manuscript; no accuracy numbers, confidence intervals, significance tests, or interpretability measurements are reported. Since this claim is the paper's main contribution, the manuscript as submitted is unverifiable.
  2. [Abstract, case-study selection] The two systems LNN and LLM-SS are described as 'introduced' for this study rather than selected from a pool of established best-in-class systems; no selection criteria, external baselines, or multiple instances of each family are provided. The family-level conclusion therefore does not follow from the two-system comparison.
  3. [Abstract, supporting reason (i)] The claim that the hybrid reasoning chain is 'more interpretable' is close to being definitional, because the hybrid design externalizes symbolic reasoning by construction; without a task-specific interpretability protocol and a comparison against at least one integrative system that also produces inspectable traces, this premise has no independent empirical content.
  4. [Full text, general] The supplied full text is largely corrupted (mojibake), including the tables that appear to report experimental results; because the methods, experimental setup, and results cannot be read, I cannot verify the derivation of the proposed framework or the comparative conclusion. This is a blocking issue for any recommendation other than major revision.
minor comments (3)
  1. [Abstract] The phrase 'to the best of our knowledge, there has not been a comparison' should be supported by a literature search and citations; as written it is an unverifiable claim.
  2. [Tables 2-5] The table cells in the supplied version are unreadable; in a revised submission, each table should have clear column headers, units, sample sizes, and error bars.
  3. [Notation] If 'LNN' refers to an existing Logic Neural Network architecture, the authors should clarify what is novel in their instantiation; if it is a new model, the name is potentially confusing.

Circularity Check

1 steps flagged · score 6.0 of 10

The comparative conclusion follows from the paper's own approach definitions and case-study construction, not from independent measurement.

  1. self definitional [Abstract, paragraph 2 (approach definitions and central conclusion)]
    "the hybrid approach comprising models where a symbolic solver, separate from the neural network, performs symbolic reasoning... our analysis demonstrates that the hybrid approach is more promising for developing general logical reasoning because (i) its reasoning chain is more interpretable, and (ii) it retains the capabilities and advantages of existing LLMs."

    Reason (i) is not established by measurement; it is entailed by the definition chosen for 'hybrid': because the solver is separate and external, its reasoning chain is inspectable by construction, while the integrative definition contains symbolic reasoning inside the network, so no external chain is defined. Reason (ii) is likewise entailed by selecting LLM-SS, an LLM-based system, as the sole hybrid representative; the claim that the hybrid approach retains LLM capabilities restates the construction of the representative rather than an empirical property of the approach family. The conclusion is therefore a repackaging of the definitions and case-study choice, not an independent comparative result.

full rationale

The paper's central verdict that the hybrid approach is more promising is supported by two stated reasons, and both reasons reduce to how the approaches and representatives are defined. An external symbolic solver is definitionally separate from the neural network, so its reasoning chain is visible by construction; an LLM-Symbolic Solver is definitionally built on an LLM, so it retains LLM capabilities by construction. No independent interpretability protocol, external benchmark, or selection procedure for 'best-in-class' representatives is visible in the supplied text, and the full text is corrupted, so I cannot verify that the body supplies such evidence. The 'introduced' rather than independently selected representatives also makes the family-level generalization depend on an unvalidated input, though that concern is more a correctness risk than a circular derivation. Because the two headline reasons for preferring hybrid over integrative are definitional or constructional rather than measured, the paper exhibits partial circularity, but I do not find a self-citation chain or a fitted-parameter prediction, so the score is 6 rather than higher.

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

No equations or fitted parameters appear in the abstract, so the free-parameter count is zero. The three listed axioms are the assumptions needed to move from two case studies to a general design recommendation. The corrupted full text prevented an audit of any additional assumptions in the body.

assumptions (3)
  • domain assumption LNN and LLM-SS are representative, best-in-class domain-agnostic models for the integrative and hybrid approaches respectively.
    The general comparison conclusion depends on these two specific models standing in for their entire families. Stated in the abstract where the models are introduced.
  • domain assumption Interpretability and retention of LLM capabilities are the correct criteria for judging which approach is more promising.
    The abstract's conclusion is justified with exactly these two criteria; no independent objective metric is given.
  • domain assumption The domain-agnostic benchmarks used in the study are valid measures of general logical reasoning.
    The abstract describes the models as domain-agnostic but does not name the benchmarks; the comparison's validity depends on this unstated benchmark setup.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Comparative Study of Neurosymbolic AI Approaches to Interpretable Logical Reasoning." pith.science (2026). https://pith.science/paper/PX7JLNVP

@misc{pith2026250803366,
  author       = {Pith},
  title        = {Pith review of: A Comparative Study of Neurosymbolic AI Approaches to Interpretable Logical Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PX7JLNVP}},
  note         = {Machine review of arXiv:2508.03366}
}
read the original abstract

General logical reasoning, defined as the ability to reason deductively on domain-agnostic tasks, continues to be a challenge for large language models (LLMs). Current LLMs fail to reason deterministically and are not interpretable. As such, there has been a recent surge in interest in neurosymbolic AI, which attempts to incorporate logic into neural networks. We first identify two main neurosymbolic approaches to improving logical reasoning: (i) the integrative approach comprising models where symbolic reasoning is contained within the neural network, and (ii) the hybrid approach comprising models where a symbolic solver, separate from the neural network, performs symbolic reasoning. Both contain AI systems with promising results on domain-specific logical reasoning benchmarks. However, their performance on domain-agnostic benchmarks is understudied. To the best of our knowledge, there has not been a comparison of the contrasting approaches that answers the following question: Which approach is more promising for developing general logical reasoning? To analyze their potential, the following best-in-class domain-agnostic models are introduced: Logic Neural Network (LNN), which uses the integrative approach, and LLM-Symbolic Solver (LLM-SS), which uses the hybrid approach. Using both models as case studies and representatives of each approach, our analysis demonstrates that the hybrid approach is more promising for developing general logical reasoning because (i) its reasoning chain is more interpretable, and (ii) it retains the capabilities and advantages of existing LLMs. To support future works using the hybrid approach, we propose a generalizable framework based on LLM-SS that is modular by design, model-agnostic, domain-agnostic, and requires little to no human input.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. From Judgments to Issues: Structured Extraction of Legal Reasoning with Citation-Hallucination Control

    cs.CL 2026-07 conditional novelty 6.0 of 10

    A cost-efficient DeepSeek-V3 pipeline extracts IRAC-grounded issue-level XML from ~330k Italian tax judgments and cuts citation hallucinations from 11.7% to 0.9% via Linkoln matching, validated by two tax-law PhDs on ...

  2. Probabilistic Extension of Neuro-Symbolic AGI Robots based on Belnap's Typed Intensional FOL

    cs.AI 2026-07 reject novelty 3.0 of 10

    A theoretical paper defines global and local 'symmetry transformations' to assign probabilities to unknown sentences in a Belnap four-valued intensional logic, with no implementation or validation.

Reference graph

Works this paper leans on

39 extracted references · 17 canonical work pages · cited by 2 Pith papers

  1. [1]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023

    arXiv 2023

  2. [2]

    Uci machine learning repository, 2007

    Arthur Asuncion, David Newman, et al. Uci machine learning repository, 2007

    2007

  3. [3]

    Logic tensor networks

    Samy Badreddine, Artur d'Avila Garcez , Luciano Serafini, and Michael Spranger. Logic tensor networks. Artificial Intelligence, 303: 0 103649, 2022. ISSN 0004-3702. doi:https://doi.org/10.1016/j.artint.2021.103649

    arXiv 2022

  4. [4]

    Qwen technical report

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. Qwen technical report. arXiv preprint arXiv:2309.16609, 2023

    arXiv 2023

  5. [5]

    Truth table net: Scalable, compact & verifiable neural networks with a dual convolutional small boolean circuit networks form

    Adrien Benamira, Thomas Peyrin, Trevor Yap, Tristan Gu \'e rand, and Bryan Hooi. Truth table net: Scalable, compact & verifiable neural networks with a dual convolutional small boolean circuit networks form

    work page

  6. [6]

    Learning Symbolic Expressions via Gumbel-Max Equation Learner Networks

    Gang Chen. Learning symbolic expressions via gumbel-max equation learner networks. arXiv preprint arXiv:2012.06921, 2020

    work page Pith review arXiv 2012

  7. [7]

    Symbolic knowledge extraction and injection with sub-symbolic predictors: A systematic literature review

    Giovanni Ciatto, Federico Sabbatini, Andrea Agiollo, Matteo Magnini, and Andrea Omicini. Symbolic knowledge extraction and injection with sub-symbolic predictors: A systematic literature review. ACM Computing Surveys, 56 0 (6): 0 1--35, 2024

    work page 2024

  8. [8]

    Bridging machine learning and logical reasoning by abductive learning

    Wang-Zhou Dai, Qiuling Xu, Yang Yu, and Zhi-Hua Zhou. Bridging machine learning and logical reasoning by abductive learning. Advances in Neural Information Processing Systems, 32, 2019

    2019

Show all 39 references
  1. [9]

    Knowledge enhanced neural networks

    Alessandro Daniele and Luciano Serafini. Knowledge enhanced neural networks. In PRICAI 2019: Trends in Artificial Intelligence: 16th Pacific Rim International Conference on Artificial Intelligence, Cuvu, Yanuca Island, Fiji, August 26--30, 2019, Proceedings, Part I 16, pages 5...

    2019

  2. [10]

    Dl2: training and querying neural networks with logic

    Marc Fischer, Mislav Balunovic, Dana Drachsler-Cohen, Timon Gehr, Ce Zhang, and Martin Vechev. Dl2: training and querying neural networks with logic. In International Conference on Machine Learning, pages 1931--1941. PMLR, 2019

    1931

  3. [11]

    Artur d’Avila Garcez and Lu\' s C. Lamb. Neurosymbolic ai: the 3rd wave. Artif. Intell. Rev., 56 0 (11): 0 12387–12406, mar 2023. ISSN 0269-2821. doi:10.1007/s10462-023-10448-w. URL https://doi.org/10.1007/s10462-023-10448-w

    2023 doi

  4. [12]

    Multi-shot asp solving with clingo

    Martin Gebser, Roland Kaminski, Benjamin Kaufmann, and Torsten Schaub. Multi-shot asp solving with clingo. Theory and Practice of Logic Programming, 19 0 (1): 0 27--82, 2019

    2019

  5. [13]

    Did aristotle use a laptop? a question answering benchmark with implicit reasoning strategies

    Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. Did aristotle use a laptop? a question answering benchmark with implicit reasoning strategies. Transactions of the Association for Computational Linguistics, 9: 0 346--361, 2021

    2021

  6. [14]

    Lubm: A benchmark for owl knowledge base systems

    Yuanbo Guo, Zhengxiang Pan, and Jeff Heflin. Lubm: A benchmark for owl knowledge base systems. Journal of Web Semantics, 3 0 (2-3): 0 158--182, 2005

    2005

  7. [15]

    Folio: Natural language reasoning with first-order logic

    Simeng Han, Hailey Schoelkopf, Yilun Zhao, Zhenting Qi, Martin Riddell, Wenfei Zhou, James Coady, David Peng, Yujie Qiao, Luke Benson, et al. Folio: Natural language reasoning with first-order logic. arXiv preprint arXiv:2209.00840, 2022

    2022 arXiv

  8. [16]

    Measuring massive multitask language understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300, 2020

    2009 arXiv

  9. [17]

    Towards reasoning in large language models: A survey

    Jie Huang and Kevin Chen-Chuan Chang. Towards reasoning in large language models: A survey. arXiv preprint arXiv:2212.10403, 2022

    2022 arXiv

  10. [18]

    Third AI Summer, AAAI Robert S

    H Kautz. Third AI Summer, AAAI Robert S. Engelmore Memorial Lecture , 2020. URL https://www.cs.rochester.edu/u/kautz/talks/Kautz

    2020

  11. [19]

    Abstract meaning representation (amr) annotation release 3.0

    Kevin Knight, Bianca Badarau, Laura Baranescu, Claire Bonial, Madalina Bardocz, Kira Griffitt, Ulf Hermjakob, Daniel Marcu, Martha Palmer, Tim O'Gorman, et al. Abstract meaning representation (amr) annotation release 3.0. Linguistic Data Consortium, 2020., 2021

    2020

  12. [20]

    Maximum bayes smatch ensemble distillation for amr parsing

    Young-Suk Lee, Ramon Fernandez Astudillo, Thanh Lam Hoang, Tahira Naseem, Radu Florian, and Salim Roukos. Maximum bayes smatch ensemble distillation for amr parsing. arXiv preprint arXiv:2112.07790, 2021

    2021 arXiv

  13. [21]

    Answer set programming, volume 3

    Vladimir Lifschitz. Answer set programming, volume 3. Springer Heidelberg, 2019

    2019

  14. [22]

    Foundations of logic programming

    John W Lloyd. Foundations of logic programming. Springer Science & Business Media, 2012

    2012

  15. [23]

    Faithful chain-of-thought reasoning

    Qing Lyu, Shreya Havaldar, Adam Stein, Li Zhang, Delip Rao, Eric Wong, Marianna Apidianaki, and Chris Callison-Burch. Faithful chain-of-thought reasoning. In The 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapt...

    2023

  16. [24]

    Deepproblog: Neural probabilistic logic programming

    Robin Manhaeve, Sebastijan Dumancic, Angelika Kimmig, Thomas Demeester, and Luc De Raedt. Deepproblog: Neural probabilistic logic programming. Advances in neural information processing systems, 31, 2018

    2018

  17. [25]

    Automated theorem provers help improve large language model reasoning

    Lachlan McGinness and Peter Baumgartner. Automated theorem provers help improve large language model reasoning. In Proceedings of 25th Conference on Logic for Programming, Artificial Intelligence and Reasoning, volume 100, pages 51--69, 2024

    2024

  18. [26]

    Unifying large language models and knowledge graphs: A roadmap

    Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu. Unifying large language models and knowledge graphs: A roadmap. IEEE Transactions on Knowledge and Data Engineering, 36 0 (7): 0 3580--3599, 2024. doi:10.1109/TKDE.2024.3352100

    2024

  19. [27]

    Deep differentiable logic gate networks

    Felix Petersen, Christian Borgelt, Hilde Kuehne, and Oliver Deussen. Deep differentiable logic gate networks. Advances in Neural Information Processing Systems, 35: 0 2006--2018, 2022

    2006

  20. [28]

    Introduction to Logic: Propositional Logic

    Howard Pospesel. Introduction to Logic: Propositional Logic. Prentice-Hall, Englewood Cliffs, NJ, USA, 1974

    1974

  21. [29]

    Logical neural networks

    Ryan Riegel, Alexander Gray, Francois Luus, Naweed Khan, Ndivhuwo Makondo, Ismail Yunus Akhalwaya, Haifeng Qian, Ronald Fagin, Francisco Barahona, Udit Sharma, et al. Logical neural networks. arXiv preprint arXiv:2006.13155, 2020

    2006 arXiv

  22. [30]

    Stepgame: A new benchmark for robust multi-hop spatial reasoning in texts

    Zhengxiang Shi, Qiang Zhang, and Aldo Lipani. Stepgame: A new benchmark for robust multi-hop spatial reasoning in texts. In Proceedings of the AAAI conference on artificial intelligence, volume 36, pages 11321--11329, 2022

    2022

  23. [31]

    Mastering the game of go with deep neural networks and tree search

    David Silver, Aja Huang, Chris J Maddison, Arthur Guez, Laurent Sifre, George Van Den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lanctot, et al. Mastering the game of go with deep neural networks and tree search. nature, 529 0 (7587): 0 484-...

    2016

  24. [32]

    Clutrr: A diagnostic benchmark for inductive reasoning from text

    Koustuv Sinha, Shagun Sodhani, Jin Dong, Joelle Pineau, and William L Hamilton. Clutrr: A diagnostic benchmark for inductive reasoning from text. arXiv preprint arXiv:1908.06177, 2019

    1908 arXiv

  25. [33]

    Llama 2: Open foundation and fine-tuned chat models

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  26. [34]

    Solving olympiad geometry without human demonstrations

    Trieu H Trinh, Yuhuai Wu, Quoc V Le, He He, and Thang Luong. Solving olympiad geometry without human demonstrations. Nature, 625 0 (7995): 0 476--482, 2024

    2024

  27. [35]

    Analyzing differentiable fuzzy logic operators

    Emile van Krieken, Erman Acar, and Frank van Harmelen. Analyzing differentiable fuzzy logic operators. Artificial Intelligence, 302: 0 103602, 2022

    2022

  28. [36]

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

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35: 0 24824--24837, 2022

    2022

  29. [37]

    Coupling large language models with logic programming for robust and general reasoning from text

    Zhun Yang, Adam Ishay, and Joohyung Lee. Coupling large language models with logic programming for robust and general reasoning from text. arXiv preprint arXiv:2307.07696, 2023

    2023 arXiv

  30. [38]

    Improved logical reasoning of language models via differentiable symbolic programming

    Hanlin Zhang, Jiani Huang, Ziyang Li, Mayur Naik, and Eric Xing. Improved logical reasoning of language models via differentiable symbolic programming. arXiv preprint arXiv:2305.03742, 2023

    2023 arXiv

  31. [39]

    Structure-aware fine-tuning of sequence-to-sequence transformers for transition-based amr parsing

    Jiawei Zhou, Tahira Naseem, Ram \'o n Fernandez Astudillo, Young-Suk Lee, Radu Florian, and Salim Roukos. Structure-aware fine-tuning of sequence-to-sequence transformers for transition-based amr parsing. arXiv preprint arXiv:2110.15534, 2021

    arXiv

Pith tools

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