GPT-4 Can't Reason
Open accessAbstract
GPT-4 was released in March 2023 to wide acclaim, marking a very substantial improvement across the board over GPT-3.5 (OpenAI's previously best model, which had powered the initial release of ChatGPT). Despite the genuinely impressive improvement, however, there are good reasons to be highly skeptical of GPT-4's ability to reason. This position paper discusses the nature of reasoning; criticizes the current formulation of reasoning problems in the NLP community and the way in which the reasoning performance of LLMs is currently evaluated; introduces a collection of 21 diverse reasoning problems; and performs a detailed qualitative analysis of GPT-4's performance on these problems. Based on the results of this analysis, the paper argues that, despite the occasional flashes of analytical brilliance, GPT-4 at present is utterly incapable of reasoning.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Even the most capable models at present fail at basic arithmetic problems such as multiplying two four-digit numbers when directly asked for the answer as opposed to writing code.
“It has been, for instance, repeatedly demonstrated that even the most capable models at present fail at basic arithmetic problems such as multiplying two four-digit numbers (22) when directly asked for the answer as opposed to writing code.”
From How should the advancement of large language models affect the practice of science? (2025) · cites this paper as [22]
Deficiencies of LLM generalization have been identified in multiple straightforward tasks, including simple mathematical reasoning, and in planning, even with current state-of-the-art models.
“Deficiencies of LLM generalization have been identified in multiple straightforward tasks18,19,64, including simple mathematical reasoning65, and in planning, even with current state-of-the-art models62.”
From Limitations of large language models in clinical problem-solving arising from inflexible reasoning (2025) · cites this paper as [64]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference
McCoy, R. T., Pavlick, E., & Linzen, T. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1334 · Cited by 944 · see citing works
Barwise, J., & Perry, J. (2004) · doi:10.1093/oso/9780195136975.003.0015 · Cited by 682 · see citing works
ReAct: Synergizing Reasoning and Acting in Language Models
Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., & Cao, Y. (2022) · arXiv · doi:10.48550/arxiv.2210.03629 · Cited by 608 · see citing works
Probing Neural Network Comprehension of Natural Language Arguments
Niven, T., & Kao, H.-Y. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1459 · Cited by 395 · see citing works
Evaluating the logical reasoning ability of chatgpt and GPT-4
Liu, H., Ning, R., Teng, Z., Liu, J., Zhou, Q., & Zhang, Y. (2023) · arXiv · doi:10.48550/arxiv.2304.03439 · Cited by 100 · see citing works
On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspective
Wang, J., Hu, X., Hou, W., Chen, H., Zheng, R., Wang, Y., Yang, L., Huang, H., Ye, W., Geng, X., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.12095 · Cited by 86 · see citing works
Karpas, E., Abend, O., Belinkov, Y., Lenz, B., Lieber, O., Ratner, N., Shoham, Y., Bata, H., Levine, Y., Leyton-Brown, K., et al. (2022) · arXiv · doi:10.48550/arxiv.2205.00445 · Cited by 49 · see citing works
Temporal Reasoning Problems and Algorithms for Solving Them: literature survey
Planken, L. (2008) · Research Repository (Delft University of Technology) · Cited by 4 · see citing works
Flagged first, then by citations
1 of the 9 references couldn’t be matched to a known paper.
Research works printed but not linked: 2
- Arkoudas, K. and Musser, D., Fundamental Proof Methods in Computer Science, MIT Press, 2017.
- Johnson-Laird, P.N., How We Reason, Oxford University Press, 2006.
Cited by
Sorted by citations
Spennemann, D. H. R. (2023) · Knowledge · doi:10.3390/knowledge3030032 · Cited by 93 · see citing works
Limitations of large language models in clinical problem-solving arising from inflexible reasoning
Kim, J., Podlasek, A., Shidara, K., Liu, F., Alaa, A., & Bernardo, D. (2025) · Scientific Reports · doi:10.1038/s41598-025-22940-0 · Cited by 78 · see citing works
How should the advancement of large language models affect the practice of science?
Binz, M., Alaniz, S., Roskies, A., Aczel, B., Bergstrom, C. T., Allen, C., Schad, D., Wulff, D., West, J. D., Zhang, Q., et al. (2025) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2401227121 · Cited by 74 · see citing works
From concept to manufacturing: evaluating vision-language models for engineering design
Picard, C., Edwards, K. M., Doris, A. C., Man, B., Giannone, G., Alam, M. F., & Ahmed, F. (2025) · Artificial Intelligence Review · doi:10.1007/s10462-025-11290-y · Cited by 57 · see citing works
Strategic behavior of large language models and the role of game structure versus contextual framing
Lorè, N., & Heydari, B. (2024) · Scientific Reports · doi:10.1038/s41598-024-69032-z · Cited by 28 · see citing works
Can ChatGPT-4o Really Pass Medical Science Exams? A Pragmatic Analysis Using Novel Questions
Newton, P. M., Summers, C. J., Zaheer, U., Xiromeriti, M., Stokes, J. R., Bhangu, J. S., Roome, E. G., Roberts-Phillips, A., Mazaheri-Asadi, D., Jones, C. D., et al. (2025) · Medical Science Educator · doi:10.1007/s40670-025-02293-z · Cited by 18 · see citing works
Tapia-Mandiola, S., & Araya, R. (2024) · AI · doi:10.3390/ai5040093 · Cited by 11 · see citing works
Epistemology in the Age of Large Language Models
Mugleston, J. D., Truong, V. H., Kuang, C., Sibiya, L., & Myung, J. (2025) · Knowledge · doi:10.3390/knowledge5010003 · Cited by 8 · see citing works
Diversity of Thought Elicits Stronger Reasoning Capabilities in Multi-Agent Debate FrameworksPreprint
Hegazy, M. (2024) · Qeios · doi:10.32388/3y8v71 · Cited by 5 · see citing works
LLMs and the Logical Space of Reasons
Sambrotta, M. (2025) · Minds and Machines · doi:10.1007/s11023-025-09751-y · Cited by 5 · see citing works
Vervoort, L., & Nikolaev, V. (2025) · Journal for General Philosophy of Science · doi:10.1007/s10838-025-09735-7 · Cited by 4 · see citing works
Chinese Chat Room: AI Hallucinations, Epistemology and Cognition
Šekrst, K. (2024) · Studies in Logic Grammar and Rhetoric · doi:10.2478/slgr-2024-0029 · Cited by 4 · see citing works
Creating certainty where there is none: Artificial intelligence as political concept
Garibay-Petersen, C., Lorimer, M., & Menzat, B. (2025) · Big Data & Society · doi:10.1177/20539517251396079 · Cited by 7 · see citing works
Fin-ALICE: Artificial Linguistic Intelligence Causal Econometrics
McCarthy, S., & Alaghband, G. (2024) · Journal of risk and financial management · doi:10.3390/jrfm17120537 · Cited by 4 · see citing works
Strübbe, S., Grünwald, A. T. D., Sidorenko, I. N., & Lampe, R. (2024) · Applied Sciences · doi:10.3390/app15010087 · Cited by 3 · see citing works
Illusory Arguments by Artificial Agents: Pernicious Legacy of the Sophists
Clark, M. H., & Bringsjord, S. (2024) · Humanities · doi:10.3390/h13030082 · Cited by 2 · see citing works
On using large language models to support social research for climate action
Nabavi, E., Maier, H. R., Razavi, S., Hindes, A., Howden, M. S., Grant, W. J., & Raman, S. (2026) · npj Climate Action · doi:10.1038/s44168-026-00375-1 · Cited by 2 · see citing works
Generative Künstliche Intelligenz
Buder, J., Lindner, M. A., Oestermeier, U., Huff, M., Gerjets, P. H., Utz, S., & Creß, U. (2024) · Psychologische Rundschau · doi:10.1026/0033-3042/a000699 · Cited by 2 · see citing works
Hernández-Espinosa, A., Ozelim, L., Abrahão, F. S., & Zenil, H. (2026) · Nature Communications · doi:10.1038/s41467-026-73289-5 · Cited by 1 · see citing works
Каково это — быть рассуждающей языковой моделью?
Д.В., З. (2024) · Логико-философские штудии · doi:10.52119/lphs.2024.51.82.003 · Cited by 1 · see citing works
Prolegomenon for a Family of Theorems Regarding Trustworthiness in Autonomous Artificial Agents
Bringsjord, S., Govindarajulu, N. S., & Giancola, M. (2024) · Studies in computational intelligence · doi:10.1007/978-3-031-55817-7_9
GenAI & The Dual-Deception Process in Web Science
Bringsjord, A. (2025) · ACM Web Science Conference (WebSci) · doi:10.1145/3720554.3735862
Bringsjord, S., & Govindarajulu, N. S. (2026) · Studies in computational intelligence · doi:10.1007/978-3-032-06573-5_10
Vervoort, L. (2026) · Springer eBooks · doi:10.1007/978-3-032-17756-8_8
The Method of Synthetic Philosophy. Extending the Web of Knowledge Through Problem Solving
Vervoort, L. (2026) · Springer eBooks · doi:10.1007/978-3-032-17756-8_1
Multi-model relevance-guided reverse reasoning: improving chain-of-thought with small models
Hu, B., Fu, Y., & Ma, L. (2026) · The Journal of Supercomputing · doi:10.1007/s11227-026-08783-0
- Authors
- Arkoudas, K.
- Corresponding author
- Arkoudas, Konstantine
- Venue
- Preprints.org
- Publisher
- MDPI AG
- Published
- 2023-08-02
- OpenAlex
- W4385495347
- Cited by
- 26
- References
- 9
- Citations by year
- 2026: 6 · 2025: 9 · 2024: 10 · 2023: 1
- Type
- posted content
- Language
- en
- Open access
- green
- doi.org
- Locations
- Preprints.org (acceptedVersion, cc-by)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Logic, Reasoning, and Knowledge
- Keywords
- GPT-4, large language models, reasoning, reasoning problems, reasoning evaluation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
