ChatGPT Health

Btw, context caching can overcome this, e.g. https://ai.google.dev/gemini-api/docs/caching . However, this means it needs to persist the (large) state in the server side, so it may have costs associated to it.

The anecdote in question is not about mis-diagnosis, it's about a delayed diagnosis. And yeah, the inquiry sent a doctor down three paths, one of which led to a diagnosis, so let's be clear: no, the doctor didn't get it completely on their own, and: ChatGPT was, at best, 33% correct.

The biggest problem in medicine right now (that's creating a lot of the issues people have with it I'd claim) is twofold:

- Engaging with it is expensive, which raises the expectations of quality of service substantially on the part of the patients and their families

- Virtually every doctor I've ever talked to complains about the same things: insufficient time to give proper care and attention to patients, and the overbearingness of insurance companies. And these two lead into each other: so much of your doc's time is spent documenting your case. Basically every hour of patient work on their part requires a second hour of charting to document it. Imagine having to write documentation for an hour for every hour of coding you did, I bet you'd be behind a lot too. Add to it how overworked and stretched every medical profession is from nursing to doctors themselves, and you have a recipe for a really shitty experience on the part of the patients, a lot of whom, like doctors, spend an inordinate amount of time fighting with insurance companies.

> How often is "user research" helping or hurting the process of getting good health outcomes?

Depends on the quality of the research. In the case of this anecdote, I would say middling. I would also say though if the anecdotes of numerous medical professionals I've heard speak on the topic are to be believed, this is an outlier in regard to it actually being good. The majority of "patient research" that shows up is new parents upset about a vaccine schedule they don't understand, and half-baked conspiracy theories from Facebook. Often both at once.

That said, any professional, doctors included, can benefit from more information from whomever they're serving. I have a great relationship with my mechanic because by the time I take my car to him, I've already ruled out a bunch of obvious stuff, and I arrive with detailed notes on what I've done, what I've tried, what I've replaced, and most importantly: I'm honest about it. I point exactly where my knowledge on the vehicle ends, and hope he can fill in the blanks, or at least he'll know where to start poking. The problem there is the vast majority of the time, people don't approach doctors as "professionals who know more than me who can help me solve a problem," they approach them as ideological enemies and/or gatekeepers of whatever they think they need, which isn't helpful and creates conflict.

> Are there medical boards that are sending PSAs to help doctors improve common mis-diagnosis?

Doctors have shitloads of journals and reading materials that are good for them to go through, which also factors into their overworked-ness but nevertheless; yes.

> Whats the role of LLMs in all of this?

Honestly I see a lot of applications of them in the insurance side of things, unless we wanted to do something cool and like, get a decent healthcare system going.

A doctor can also have an in-the-moment negatively impactful context: depression, exhaustion, or any number of life events going on, all of which can drastically impact their performance. Doctors get depressed like everybody else. They can care less due to something affecting them. These are not problems a good LLM has.