Performance of ChatGPT on Ophthalmology-Related Questions Across Various Examination Levels: Observational Study
Bibliographic record
Abstract
BACKGROUND: ChatGPT and language learning models have gained attention recently for their ability to answer questions on various examinations across various disciplines. The question of whether ChatGPT could be used to aid in medical education is yet to be answered, particularly in the field of ophthalmology. OBJECTIVE: The aim of this study is to assess the ability of ChatGPT-3.5 (GPT-3.5) and ChatGPT-4.0 (GPT-4.0) to answer ophthalmology-related questions across different levels of ophthalmology training. METHODS: Questions from the United States Medical Licensing Examination (USMLE) steps 1 (n=44), 2 (n=60), and 3 (n=28) were extracted from AMBOSS, and 248 questions (64 easy, 122 medium, and 62 difficult questions) were extracted from the book, Ophthalmology Board Review Q&A, for the Ophthalmic Knowledge Assessment Program and the Board of Ophthalmology (OB) Written Qualifying Examination (WQE). Questions were prompted identically and inputted to GPT-3.5 and GPT-4.0. RESULTS: GPT-3.5 achieved a total of 55% (n=210) of correct answers, while GPT-4.0 achieved a total of 70% (n=270) of correct answers. GPT-3.5 answered 75% (n=33) of questions correctly in USMLE step 1, 73.33% (n=44) in USMLE step 2, 60.71% (n=17) in USMLE step 3, and 46.77% (n=116) in the OB-WQE. GPT-4.0 answered 70.45% (n=31) of questions correctly in USMLE step 1, 90.32% (n=56) in USMLE step 2, 96.43% (n=27) in USMLE step 3, and 62.90% (n=156) in the OB-WQE. GPT-3.5 performed poorer as examination levels advanced (P<.001), while GPT-4.0 performed better on USMLE steps 2 and 3 and worse on USMLE step 1 and the OB-WQE (P<.001). The coefficient of correlation (r) between ChatGPT answering correctly and human users answering correctly was 0.21 (P=.01) for GPT-3.5 as compared to -0.31 (P<.001) for GPT-4.0. GPT-3.5 performed similarly across difficulty levels, while GPT-4.0 performed more poorly with an increase in the difficulty level. Both GPT models performed significantly better on certain topics than on others. CONCLUSIONS: ChatGPT is far from being considered a part of mainstream medical education. Future models with higher accuracy are needed for the platform to be effective in medical education.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.008 | 0.062 |
| Meta-epidemiology (narrow) | 0.000 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.003 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".