Comparative performance of humans versus GPT-4.0 and GPT-3.5 in the self-assessment program of American Academy of Ophthalmology
Open accessThe DOI matches its Crossref record; 9 of the 9 references it cites were found in a registry or an index.
Abstract
To compare the performance of humans, GPT-4.0 and GPT-3.5 in answering multiple-choice questions from the American Academy of Ophthalmology (AAO) Basic and Clinical Science Course (BCSC) self-assessment program, available at https://www.aao.org/education/self-assessments . In June 2023, text-based multiple-choice questions were submitted to GPT-4.0 and GPT-3.5. The AAO provides the percentage of humans who selected the correct answer, which was analyzed for comparison. All questions were classified by 10 subspecialties and 3 practice areas (diagnostics/clinics, medical treatment, surgery). Out of 1023 questions, GPT-4.0 achieved the best score (82.4%), followed by humans (75.7%) and GPT-3.5 (65.9%), with significant difference in accuracy rates (always P < 0.0001). Both GPT-4.0 and GPT-3.5 showed the worst results in surgery-related questions (74.6% and 57.0% respectively). For difficult questions (answered incorrectly by > 50% of humans), both GPT models favorably compared to humans, without reaching significancy. The word count for answers provided by GPT-4.0 was significantly lower than those produced by GPT-3.5 (160 ± 56 and 206 ± 77 respectively, P < 0.0001); however, incorrect responses were longer (P < 0.02). GPT-4.0 represented a substantial improvement over GPT-3.5, achieving better performance than humans in an AAO BCSC self-assessment test. However, ChatGPT is still limited by inconsistency across different practice areas, especially when it comes to surgery.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Sallam, M. (2023) · Healthcare · doi:10.3390/healthcare11060887 · Cited by 3,126
Gilson, A., Safranek, C. W., Huang, T., Socrates, V., Chi, L., Taylor, R. A., & Chartash, D. (2023) · JMIR Medical Education · doi:10.2196/45312 · Cited by 2,078
Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine
Lee, P., Bubeck, S., & Petro, J. (2023) · New England Journal of Medicine · doi:10.1056/nejmsr2214184 · Cited by 1,738
Artificial hallucinations in ChatGPT: implications in scientific writing
Alkaissi, H., & McFarlane, S. I. (2023) · Cureus · doi:10.7759/cureus.35179 · Cited by 1,540 · cited 1×
“Finally, AI can generate hallucinations, consisting of completely fabricated information, factual inaccuracies, logical inconsistencies or nonsensical responses 13,14 .”
BackgroundDave, T., Athaluri, S. A., & Singh, S. (2023) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2023.1169595 · Cited by 1,204
Gao, C. A., Howard, F. M., Markov, N. S., Dyer, E. C., Ramesh, S., Luo, Y., & Pearson, A. T. (2023) · npj Digital Medicine · doi:10.1038/s41746-023-00819-6 · Cited by 1,099
Wach, K., Duong, C. D., Ejdys, J., Kazlauskaitė, R., Korzynski, P., Mazurek, G., Paliszkiewicz, J., & Ziemba, E. (2023) · Entrepreneurial Business and Economics Review · doi:10.15678/eber.2023.110201 · Cited by 584 · cited 2×
“Finally, AI can generate hallucinations, consisting of completely fabricated information, factual inaccuracies, logical inconsistencies or nonsensical responses 13,14 .”
BackgroundPerformance of GPT-3.5 and GPT-4 on the Japanese Medical Licensing Examination: Comparison Study
Takagi, S., Watari, T., Erabi, A., & Sakaguchi, K. (2023) · JMIR Medical Education · doi:10.2196/48002 · Cited by 271
Moshirfar, M., Altaf, A. W., Stoakes, I. M., Tuttle, J. J., & Hoopes, P. C. (2023) · Cureus · doi:10.7759/cureus.40822 · Cited by 124
Cited by
Ganjavi, C., Eppler, M., O’Brien, D., Ramacciotti, L. S., Ghauri, M. S., Anderson, I., Choi, J., Dwyer, D., Stephens, C., Shi, V., et al. (2024) · PLOS Digital Health · doi:10.1371/journal.pdig.0000596 · Cited by 73
AirGPT: Pioneering the convergence of conversational AI with atmospheric science
Song, J., Ma, C., & Ran, M. (2025) · npj Climate and Atmospheric Science · doi:10.1038/s41612-025-01070-4 · Cited by 48
Haddad, F., & Saade, J. S. (2024) · JMIR Medical Education · doi:10.2196/50842 · Cited by 48
Chen, Y.-C., Lee, S.-H., Sheu, H., Lin, S.-H., Hu, C.-C., Fu, S.-C., Yang, C.-P., & Lin, Y.-C. (2025) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-025-03024-5 · Cited by 23
“Similarly, GPT-4 demonstrated superior performance in the American Academy of Ophthalmology Basic and Clinical Science Course self-assessment test [15], the Japanese Medical Licensing Examination [14], and Radiology Diagnosis Please cases [17].”
Accuracy of large language models in answering ophthalmology board-style questions: A meta-analysis
Wu, J.-H., Nishida, T., & Liu, T. Y. A. (2024) · Asia-Pacific Journal of Ophthalmology · doi:10.1016/j.apjo.2024.100106 · Cited by 16
Metadata
- Authors
- Taloni, A., Borselli, M., Scarsi, V., Rossi, C., Coco, G., Scorcia, V., & Giannaccare, G.
- Venue
- Scientific Reports
- Published
- 2023-10-29
- Volume
- 13
- Issue
- 1
- Pages
- 18562-18562
- ISSN
- 2045-2322 (ISSN-L)
- OpenAlex
- W4388007786
- Cited by
- 104
- References
- 9
- FWCI
- 3.60
- Type
- journal article
- Language
- en
- Open access
- gold
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Comparative performance of humans versus GPT-4.0 and GPT-3.5 in the self-assessment program of American Academy of Ophthalmology. https://app.cochat.ai/papers/10.1038/s41598-023-45837-2
- The paper
- Taloni, A., Borselli, M., Scarsi, V., Rossi, C., Coco, G., Scorcia, V., & Giannaccare, G. (2023). Comparative performance of humans versus GPT-4.0 and GPT-3.5 in the self-assessment program of American Academy of Ophthalmology. Scientific Reports. https://doi.org/10.1038/s41598-023-45837-2
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.