Performance of ChatGPT on the Chinese Postgraduate Examination for Clinical Medicine: Survey Study
Bibliographic record
Abstract
BACKGROUND: ChatGPT, an artificial intelligence (AI) based on large-scale language models, has sparked interest in the field of health care. Nonetheless, the capabilities of AI in text comprehension and generation are constrained by the quality and volume of available training data for a specific language, and the performance of AI across different languages requires further investigation. While AI harbors substantial potential in medicine, it is imperative to tackle challenges such as the formulation of clinical care standards; facilitating cultural transitions in medical education and practice; and managing ethical issues including data privacy, consent, and bias. OBJECTIVE: The study aimed to evaluate ChatGPT's performance in processing Chinese Postgraduate Examination for Clinical Medicine questions, assess its clinical reasoning ability, investigate potential limitations with the Chinese language, and explore its potential as a valuable tool for medical professionals in the Chinese context. METHODS: A data set of Chinese Postgraduate Examination for Clinical Medicine questions was used to assess the effectiveness of ChatGPT's (version 3.5) medical knowledge in the Chinese language, which has a data set of 165 medical questions that were divided into three categories: (1) common questions (n=90) assessing basic medical knowledge, (2) case analysis questions (n=45) focusing on clinical decision-making through patient case evaluations, and (3) multichoice questions (n=30) requiring the selection of multiple correct answers. First of all, we assessed whether ChatGPT could meet the stringent cutoff score defined by the government agency, which requires a performance within the top 20% of candidates. Additionally, in our evaluation of ChatGPT's performance on both original and encoded medical questions, 3 primary indicators were used: accuracy, concordance (which validates the answer), and the frequency of insights. RESULTS: Our evaluation revealed that ChatGPT scored 153.5 out of 300 for original questions in Chinese, which signifies the minimum score set to ensure that at least 20% more candidates pass than the enrollment quota. However, ChatGPT had low accuracy in answering open-ended medical questions, with only 31.5% total accuracy. The accuracy for common questions, multichoice questions, and case analysis questions was 42%, 37%, and 17%, respectively. ChatGPT achieved a 90% concordance across all questions. Among correct responses, the concordance was 100%, significantly exceeding that of incorrect responses (n=57, 50%; P<.001). ChatGPT provided innovative insights for 80% (n=132) of all questions, with an average of 2.95 insights per accurate response. CONCLUSIONS: Although ChatGPT surpassed the passing threshold for the Chinese Postgraduate Examination for Clinical Medicine, its performance in answering open-ended medical questions was suboptimal. Nonetheless, ChatGPT exhibited high internal concordance and the ability to generate multiple insights in the Chinese language. Future research should investigate the language-based discrepancies in ChatGPT's performance within the health care context.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.007 | 0.016 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 teacher head, 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".