Cool! ChatGPT Almost Graduates The Hardest Medical Licensing Test In The US

OpenAI's ChatGPT shows its ability to follow the US Medical Licensing Exam (USMLE) (photo: dock. unsplash)

JAKARTA - Once again, OpenAI's ChatGPT Artificial Intelligence (AI) chatbot shows its ability by following the US Medical Licensing Exam (USMLE). Even though it did not pass, this tool was declared almost successful in completing 75 percent in three tests.

USMLE, is known as a test with a level of difficulty and usually takes about 300 to 400 hours to complete it.

The medical examination covers everything from basic science concepts to bioethics. USMLE is actually three tests that are combined into one, and competency.

Meanwhile, ChatGPT is an AI type known as a large language model or LM. In particular, this LLM is directed at written response, through large sample text and some intelligent algorithms.

They can make predictions about which words should be combined in a sentence, such as the brothers of the phone's predictive text function.

Researchers from startup Ansible Health tested ChatGPT using an example of a question from USMLE, after examining the answer was not available on Google, so they knew ChatGPT would generate a new response based on trained data.

As a result, ChatGPT scored between 52.4 percent and 75 percent in three exams (the graduation rate is usually around 60 percent). In its 88.9 percent response, it produced at least one significant insight, described as something new, unclear, and clinically valid by the researchers.

"ChatGPT is carried out on or approaching the graduation threshold for all three exams without training or special reinforcement. In addition, ChatGPT shows high-level suitability and insight in its explanation," the researchers said in a paper published in the journal PLOS Digital Health, quoted from Science Alert, Monday, February 13.

ChatGPT also proved to be very consistent in its answer and was even able to give the reasons behind any response. It also beat the 50.3 percent accuracy rate of PubMedGPT, a bot specializedally trained on medical literature.

It should be noted that the information that has been trained by ChatGPT will include inaccuracies, and this tool will not replace medical professionals at any time in the future.