Exploring AI Hallucinations of ChatGPT
Bibliographic record
Abstract
INTRODUCTION: Large language model-based generative AI tools, such as the Chat Generative Pre-trained Transformer (ChatGPT) platform, have been used to assist with writing academic manuscripts. Little is known about ChatGPT's ability to accurately cite relevant references in health care simulation-related scholarly manuscripts. In this study, we sought to: (1) determine the reference accuracy and citation relevance among health care simulation debriefing articles generated by 2 different models of ChatGPT and (2) determine if ChatGPT models can be trained with specific prompts to improve reference accuracy and citation relevance. METHODS: The ChatGPT-4 and ChatGPT o1 models were asked to generate scholarly articles with appropriate references based upon three different article titles about health care simulation debriefing. Five articles with references were generated for each article title-3 ChatGPT-4 training conditions and 2 ChatGPT o1 training conditions. Each article was assessed independently by 2 blinded reviewers for reference accuracy and citation relevance. RESULTS: Fifteen articles were generated in total: 9 articles by ChatGPT-4 and 6 articles by ChatGPT o1. A total of 60.4% of the 303 references generated across 5 training conditions were classified as accurate, with no significant difference in reference accuracy between the 5 conditions. A total of 22.2% of the 451 citations were classified as highly relevant, with no significant difference in citation relevance across the 5 conditions. CONCLUSIONS: Among debriefing articles generated by ChatGPT-4 and ChatGPT o1, both ChatGPT models are unreliable with respect to reference accuracy and citation relevance. Reference accuracy and citation relevance for debriefing articles do not improve even with some degree of training built into ChatGPT prompts.
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 | 0.003 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.000 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| 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".