And Brian, since you were so instrumental in that analysis.

Yeah, I think an important thing that I liked when starting this project is we weren't pitting doctor versus doctor with AI.

We were trying to test two specific workflows, right? So the unit of analysis or our intervention wasn't just like we're throwing AI into the system.

We thought really deeply, and Selina and Eric thought really deeply about, you know, an AI first workflow where AI gets all the information and goes forward versus an AI second opinion workflow where the doctor themselves gets all the information, fills it out, and then an AI comes in and checks the work.

And I think thinking about thinks from a workflow level was important because you can use AI in a ton of different ways.

And the previous paper showed that there are pitfalls in just giving a doctor chat GPT or a chat box and saying like, go at it, it's super helpful, I promise.

Thinking deeply about the tendencies and the nuance of those workflows and how to use them optimally, I think was kind of the secret sauce of this.

way to put it.