Fig 4. Examples of potentially outdated or critically incorrect outputs from the model in current medical contexts.

A: A question on the primary treatment for hyperventilation syndrome in the emergency department. The suggestion of paper bag method for raising the carbon-dioxide concentration in the blood has been commonly used in the past, but it is not always the first choice, as it can worsen symptoms in certain patients with secondary hyperventilation, e.g., those with lung diseases causing low blood oxygen levels. In such answers, it seems that outdated, traditional information can prevail over the latest information, especially if it has been a standard practice over a period and related information is widely available on the Internet. B: A question on the initial outpatient treatment for a type-2 diabetes patient with poor control and combined diabetic retinopathy and neuropathy. The long-term treatment goal for diabetes is strict blood sugar control, but in this case, strict blood sugar control with sulfonylurea drugs during the initial treatment may aggravate the risk of diabetic retinopathy, raising strong concerns on GPT-4 answer. (Note: We re-inputted the prompts to obtain the above figures, and the text-contexts of the explanations from GPT-4 differed slightly from that presented in S1 Data, but their meanings are consistent).