Does ChatGPT Accurately Cite Scientific Sources?
BACKGROUND AND PURPOSE:
- ChatGPT performs well when generating new content, but performs poorly when providing scientific references
- Chen and Chen (JAMA Network Open, 2023) quantified ChatGPT’s citation error rate
METHODS:
- Study design
- The GPT-3.5 and GPT-4 models were used
- ChatGPT was asked about specific learning health systems topics followed by citation request
- E.g., “Machine learning can use EHR data. Provide 8 journal articles for stroke risk prediction models using EHR data”
- Cited journal articles were verified using Google Scholar
- To determine a reliable error rate, over 300 article references were produced on the LHS topics
- The error rate between the GPT-4 and GPT-3.5 models was compared using the Fisher exact test
- Primary outcome
- Error rate in citing scientific articles
RESULTS:
- When asked to provide citations, the error rate was significantly higher for GPT-3.5, compared to GPT-4 (P<0.001), but both had high error rates
- GPT-3.5: 98.1% (95% CI, 94.7 to 99.6)
- GPT-4: 20.6% (95% CI, 15.8 to 26.1)
- Narrower topics tended to have more fake articles than broader topics
CONCLUSION:
- GPT-4 could be used to provide citations regarding new learning health systems education and training materials, provided they are verified by humans
- The authors state
When asked why it returned fake references, ChatGPT explained that the training data may be unreliable, or the model may not be able to distinguish between reliable and unreliable sources
Learn More – Primary Sources:
Accuracy of Chatbots in Citing Journal Articles
Want to share this with your colleagues?
SPECIALTY AREAS
- Alerts
- Allergy And Immunology
- Cancer Screening
- Cardiology
- Cervical Cancer Screening
- Dermatology
- Diabetes
- Endocrine
- ENT
- Evidence Matters
- General Internal Medicine
- Genetics
- Geriatrics
- GI
- GU
- Hematology
- ID
- Medical Legal
- Mental Health
- MSK
- Nephrology
- Neurology
- PcMED Connect
- PrEP for Patients
- PrEP for Physicians
- Preventive Medicine
- Pulmonary
- Rheumatology
- Vaccinations
- Women's Health
- Your Practice