In this small but interesting study chaggpt On Its Own did much better on diagnosis than doctors with no access to chatgpt AND much better than doctors who did have access.
The doctors with access simply didn't trust chatgpt results - they preferred their own diagnosis
There're two sides to AI. You can OVERTRUST it, giving it jobs it can't do/shouldn't do. But you can also UNDERTRUST it - refusing to use it when it would be useful to do so...
