> My local model never says no.
> I don't get to hear that it's not a doctor and I should call 911.
> I don't get to hear that it's not a doctor and I should call 911.
Anthropic has a significant interest in reducing liability, which in this case means mostly blocking this use case.
Unfortunately, it's fairly tough for someone to determine if they are themselves a reasonable person.
A model that only ever says yes is a cognitohazard, IMO. Like being a billionaire without the money to insulate you from the consequences.
Sometimes, the requested information should be accompanied with a note like that.
Yes. I've also heard the cost (death) of people falling into pseudosciency rabbit holes.