#ChatGPT is still very misleading and untrustworthy when citing medical research.
https://link.springer.com/article/10.1007/s11739-023-03484-5
"We asked ChatGPT (free version 3.5) to provide us with a list of 10 references for 10 questions in internal #medicine…Of the 100 references generated by ChatGPT, 34 were correct, 40 partially correct, and 26 incorrect."
Since things are (or might be) changing fast, note that the tests were run on September 15, 2023. The authors share all their data.
Follow
@petersuber No kidding. When you use a screwdriver as a hammer, you get similar results. AI just isn't a fact machine and we need to stop pretending like it is.
