Abstract
ABSTRACT
Introduction
AI chatbots can provide personalised health information, solve examinations, and answer specialised questions. Ensuring the accuracy of these tools is essential due to the potential consequences of incorrect information. This study aimed to evaluate the performance of large language models and undergraduate students in solving questions from dental examinations.
Materials and Methods
Examination questions were entered into the paid version of ChatGPT‐5 (OpenAI, San Francisco, CA, USA) and the free version of Gemini (Google, CA, USA) on the same day. Performance was assessed by success rate (accuracy) and compared with student scores. The analysis considered multiple variables, including the number of words used in the responses, the cognitive complexity level of the questions, and the specific areas of knowledge assessed. Statistical analyses were performed using IBM SPSS Statistics, version 26.0 (IBM Corp., Armonk, NY, USA), adopting a 5% significance level (
p
Results
Across all examinations, ChatGPT and Gemini showed similar performance, with consistently high accuracy rates that exceeded those of the students. For low‐complexity questions, both models achieved 100% accuracy. For medium‐complexity items, performance ranged from 90% to 100%. In high‐complexity questions, accuracy decreased, with ChatGPT reaching 78.6% and Gemini 71.4%.
Conclusion
The large language models evaluated in this study demonstrated strong performance across all assessments, consistently outperforming undergraduate dental students. Although the integration of artificial intelligence into dentistry is still evolving, its potential clinical applications and value as a tool for accessing and managing information highlight its growing relevance in dental education and practice.