I wrote about an important #AI in healthcare study published today https://erictopol.substack.com/p/when-patient-questions-are-answered
@erictopol The example you highlight is a fascinating case study.
The GPT answer is more tactful, more full of advice, and uses more social forms.
It also starts out scary, and buries the actual answer, the most important information (“Your risk level is low”) in the 3rd paragraph. All the padding is not necessarily a blessing.
On Reddit, the human answer is the one I’d probably rather have received: a frank take on the actual risk. Yes, the social forms are nice too, but if I had to choose….
@inthehands @erictopol Anecdotal, but tried the first post in the subreddit they used for the study in ChatGPT. The med pros forcefully suggested urgent care/ER while ChatGPT responded generically with "call your physician".
