[tyoma]

> Again, don't put private health information into ChatGPT. I get the temptation, but don't do it. I'm not trying to gatekeep healthcare, but we can't trust these models to count the number of b's in blueberry consistently. If we can't trust them to do something trivial like that, can we really trust them with life-critical conversations like what happens when you're in crisis or to accurately interpret a cancer screening?

I did just this during some medical emergencies recently and ChatGPT (o3 model) did a fantastic job.

It was accurately able to give differential diagnoses that the human doctors were thinking about, accurately able to predict the tests they’d run and gave me a useful questions to ask.

It was also always available, not judgmental and you could ask it to talk in depth about conditions and possibilities without it having to rush out of the room to see another patient.

[Herring]

OpenAI released this a couple months ago

https://openai.com/index/healthbench/

Give it a year and that benchmark will probably be maxed out too.