"Where does it do well? And setting that expectation in advance is really important, not only so that we could decrease the sort of automation bias and complacency, the over-trust of AI, because we have to show people where it doesn't work, but we also have to mitigate the distrust of AI. A lot of people are afraid of using AI and if they see a case that's obviously wrong, sometimes they'll throw the baby out with the bathwater. They'll throw the model out with just that one case and really there's more value. So we have to set expectations in training appropriately. The thing I'll say to everyone is that training is not about buttonology of how to use the tool. Training in AI is a clinical training. You need clinical people to talk about where it works well and where it doesn't. And I will say that in general, people who don't have literacy in AI are going to fall behind. It's going to become a patient safety competency."