BMC medical education · 2025
Assessing ChatGPT-4's performance on the US prosthodontic exam: impact of fine-tuning and contextual prompting vs. base knowledge, a cross-sectional study.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Dashti Mahmood, Khosraviani Farshad, Azimi Tara, Hefzi Delband, Ghasemi Shohreh, Fahimipour Amir, Zare Niusha, Khurshid Zohaib, Habib Syed Rashid
Аннотация
BACKGROUND: Artificial intelligence (AI), such as ChatGPT-4 from OpenAI, has the potential to transform medical education and assessment. However, its effectiveness in specialized fields like prosthodontics, especially when comparing base to fine-tuned models, remains underexplored. This study evaluates the performance of ChatGPT-4 on the US National Prosthodontic Resident Mock Exam in its base form and after fine-tuning. The aim is to determine whether fine-tuning improves the AI's accuracy in answering specialized questions.
METHODS: An official sample questions from the 2021 US National Prosthodontic Resident Mock Exam was used, obtained from the American College of Prosthodontists. A total of 150 questions were initially considered, and resources were available for 106 questions. Both the base and fine-tuned models of ChatGPT-4 were tested under simulated exam conditions. Performance was assessed by comparing correct and incorrect responses. The Chi-square test was used to analyze accuracy, with significance set at p < 0.05. The Kappa coefficient was calculated to measure agreement between the models' responses.
RESULTS: The base model of ChatGPT-4 correctly answered 62.7% of the 150 questions. For the 106 questions with resources, the fine-tuned model answered 73.6% correctly. The Chi-square test showed a significant improvement in performance after fine-tuning (p < 0.001). The Kappa coefficient was 0.39, indicating moderate agreement between the models (p < 0.001). Performance varied by topic, with lower accuracy in areas such as Implant Prosthodontics, Removable Prosthodontics, and Occlusion, though the fine-tuned model consistently outperformed the base model.
CONCLUSIONS: Fine-tuning ChatGPT-4 with specific resources significantly enhances its accuracy in answering specialized prosthodontic exam questions. While the base model provides a solid baseline, fine-tuning is essential for improving AI performance in specialized fields. However, certain topics may require more targeted training to achieve higher accuracy.
METHODS: An official sample questions from the 2021 US National Prosthodontic Resident Mock Exam was used, obtained from the American College of Prosthodontists. A total of 150 questions were initially considered, and resources were available for 106 questions. Both the base and fine-tuned models of ChatGPT-4 were tested under simulated exam conditions. Performance was assessed by comparing correct and incorrect responses. The Chi-square test was used to analyze accuracy, with significance set at p < 0.05. The Kappa coefficient was calculated to measure agreement between the models' responses.
RESULTS: The base model of ChatGPT-4 correctly answered 62.7% of the 150 questions. For the 106 questions with resources, the fine-tuned model answered 73.6% correctly. The Chi-square test showed a significant improvement in performance after fine-tuning (p < 0.001). The Kappa coefficient was 0.39, indicating moderate agreement between the models (p < 0.001). Performance varied by topic, with lower accuracy in areas such as Implant Prosthodontics, Removable Prosthodontics, and Occlusion, though the fine-tuned model consistently outperformed the base model.
CONCLUSIONS: Fine-tuning ChatGPT-4 with specific resources significantly enhances its accuracy in answering specialized prosthodontic exam questions. While the base model provides a solid baseline, fine-tuning is essential for improving AI performance in specialized fields. However, certain topics may require more targeted training to achieve higher accuracy.