Accuracy of Information and References Using ChatGPT-3 for Retrieval of Clinical Radiological Information
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Purpose: To assess the accuracy of answers provided by ChatGPT-3 when prompted with questions from the daily routine of radiologists and to evaluate the text response when ChatGPT-3 was prompted to provide references for a given answer. Methods: ChatGPT-3 (San Francisco, OpenAI) is an artificial intelligence chatbot based on a large language model (LLM) that has been designed to generate human-like text. A total of 88 questions were submitted to ChatGPT-3 using textual prompt. These 88 questions were equally dispersed across 8 subspecialty areas of radiology. The responses provided by ChatGPT-3 were assessed for correctness by cross-checking them with peer-reviewed, PubMed-listed references. In addition, the references provided by ChatGPT-3 were evaluated for authenticity. Results: A total of 59 of 88 responses (67%) to radiological questions were correct, while 29 responses (33%) had errors. Out of 343 references provided, only 124 references (36.2%) were available through internet search, while 219 references (63.8%) appeared to be generated by ChatGPT-3. When examining the 124 identified references, only 47 references (37.9%) were considered to provide enough background to correctly answer 24 questions (37.5%). Conclusion: In this pilot study, ChatGPT-3 provided correct responses to questions from the daily clinical routine of radiologists in only about two thirds, while the remainder of responses contained errors. The majority of provided references were not found and only a minority of the provided references contained the correct information to answer the question. Caution is advised when using ChatGPT-3 to retrieve radiological information.
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.
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 | 0.004 | 0.019 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| 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 it