In a written multiple-choice test, the University of Cambridge-led study discovered that OpenAI’s large language model (LLM), GPT-4, did almost as well as specialty eye specialists.
The AI model was tested against doctors at different stages of their careers, including junior doctors without a specialization and trainee and expert eye doctors. The AI model is notable for generating text based on the large amount of data it is trained on.
Each group was given dozens of situations in which patients had a certain eye condition, and they were asked to select the best course of action for treatment or diagnosis.
The exam consisted of written questions about a variety of eye conditions, such as light sensitivity, impaired vision, lesions, and itchy eyes, that were drawn from a textbook used to assess aspiring eye doctors.
