en.wikipedia.org/wiki/Looting...

I worked at Microsoft on an automated medical scribe before the current era of generative AI. I'm familiar with HIPAA and the history of these systems :) And I think there are many positive opportunities for automated scribes. But transparency, evaluation, and consent are important to me.

There are plenty of other problems, such as inaccuracy and hallucinations, which the woman in the video very rightfully highlights. She has a legitimate complaint about systems that are MUCH more common now, and that are often used without transparency or meaningful consent.

I didn't say the article was about AI? I said it was one random article that explained some limitations of HIPAA, since you brought up HIPAA as privacy protection. Those limitations are ongoing and impact all kinds of healthcare data systems, including AI systems that interact with healthcare data.

This is great and I love Grusha and Forrest! But the grading they describe is mostly based on projects and papers, which I don't believe can be fairly graded at this point. It's just too easy to generate everything.

I can’t speak to the inner workings of every AI scribe company. I know people at Abridge and hope they’re doing things the right way! But unless I have explicit proof, I assume the worst, given history. Are local models used? Where does data go? How is data connected/sold? What specific evaluation?

A random popular article explaining some of the issues: www.scientificamerican.com/article/how-...

No, I never mentioned HIPAA. It’s very far from perfect.

I wrote that very confusingly, oops. I’m currently teaching a small seminar and have heard lots of good advice from others about to run such seminars in this day and age. But next semester I’ll be teaching a larger lecture course and am not sure how to structure things to support learning given AI.

Yeah that’s my usual inclination as well, but for discussion-based seminars, I’ve heard really good things from colleagues about printing out papers for everyone and requiring no tech in the classroom. I don’t think it makes sense for a larger lecture-based course, but I’m grasping for other ideas.

Yes I'd love to see!

Thanks but not looking for curriculum ideas! Looking for teaching methods that address how easy it is to cheat with AI.