According to a test conducted by Tiffany Kung, Victor Tseng, with coworkers at AnsibleHealth that also was published on February 9, 2023, in the fully accessible journal PLOS Digital Health, ChatGPT can score at or near the nearly 60 percent passing criterion for the United States Medical License Exam (USMLE) with responses that make clear, internal sense and offer regular insights.
 

By forecasting future word sequences, the ChatGPT languages model or new  (AI) system is created to generate writing that mimics that of a person. Unlike most other chatbots, ChatGPT cannot conduct web searches. It generates text instead based on word associations inferred by internal processes.

Kung and colleagues evaluated ChatGPT's efficiency on the USMLE, a string of three tests (Steps 1, 2CK, and 3) essential for medical licensing in the United States.

The USMLE, which med students and physicians-in-training undertake, evaluates knowledge across most medical subjects, from biochemistry to diagnostic reasoning to bioethics. After removing photo questions, the authors evaluated the algorithm on 350 of the 376 available public questions from the USMLE June 2022 release.
 

After ambiguous responses were eliminated, ChatGPT scored 52.4% and 75.0% on all three USMLE examinations. The annual passing rate is roughly 60 percent.

Also Read: Google makes its ChatGPT rival AI, "Bard," available to beta testers

ChatGPT also achieved 94.6 % concordance across all of its responses and generated at least one substantial insight (something novel, non-obvious, and clinically relevant) for 88.9 % of its responses.

OpenAI created and released ChatGPT in November 2022. It is based on OpenAI's GPT-3 family of big language models, and it has been fine-tuned (a transfer learning technique) also using supervised and reinforced learning approaches.