Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

inflecthealth.medium.com

Post reply on HN
Print view View original

211–220 of 253 posts

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#211

Post by bemusedthrow75 »

dkarras wrote: post #177
rpastuszak wrote: post #146

Earlier quoted context omitted.

> ChatGPT is excellent at writing "fluffy" pieces full of empathy, compassion, PR talk, politician speech, I think you're putting a bunch of completely different things in the same basket. Making the text more fluffy will not make it more empathetic, not only (but mainly) because there isn't anyone to empathise with the respondent. Our bullshit receptors are good spotting dishonesty, so it'll just sound cringe, putti…

>Making the text more fluffy will not make it more empathetic, Of course not, but empathy in communication (not in action) is full of fluff. It almost requires it. >Our bullshit receptors are good spotting dishonesty, so it'll just sound cringe All "genuine" affirmation for the sake of empathy sounds cringe to me. I'd rather the doctor devoted their time to doing their job well instead of trying to make someone feel…

> Of course not, but empathy in communication (not in action) is full of fluff. It almost requires it.

Empathy doesn't require fluff at all. (Think of all the short, poignant messages you've sent or received when someone is upset.)

The corporate need to not give ground on a complaint is where all that reassuring, repetitive, empty BS comes from.

> All "genuine" affirmation for the sake of empathy sounds cringe to me. I'd rather the doctor devoted their time to doing their job well instead of trying to make someone feel heard and validated - especially in an ER scenario where they are juggling critical patients. This tech can help with that.

Honestly with "for the sake of empathy" it sounds like you really don't place a high value on empathy (which is not unusual, or necessarily wrong). But if that is the case you're quite obviously not the right person to assess whether ChatGPT and the like can "help with that" in that context! :-)

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#212

Post by Spooky23 »

bohner wrote: post #85
Spooky23 wrote: post #77

Earlier quoted context omitted.

Some specialties attract different personalities. If you’ve ever been in a hospital setting around a lot of neuro residents, many of them will present… oddly. It’s a feature. Likewise, a lot of ER docs are kinda adrenaline people who are great at triage and quickly stabilizing you, but maybe not so much good at relating with people. That’s a feature — you don’t need a soft touch dealing with trauma. If ChatGPT can he…

Have worked with many poor communicators and outright idiots but there is virtually no one graduating a decent medical school with native speaker skills that cant explain in simple terms why giving more fluids to someone that is overloaded with fluid is a bad idea (oh and without using any jargon like edema or diuresis, another thing missing in this whole farce, any 1st year let alone decent attending will explain cl…

Reading the story really belies the real issue - he’s an important doctor guy who is too busy to help a couple of 70 somethings understand what he knows. I say that both in the cyclical “what an asshole” sense and in a real sense - there’s a guy having a stroke 4 bays over. That’s a sort of paradox of medicine.

I think the canned script will be handed to a less important PA or NP who will use it as a conversation guide.

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#213

Post by Spooky23 »

TeMPOraL wrote: post #144
licebmi__at__ wrote: post #119

Earlier quoted context omitted.

I disagree with the first part, mostly because I believe doctor-speak is a skill medical professionals use to deal with patients which in their desperation, think they know more than they do. Yet I fully agree with the second paragraph. The idea of the teleprompter doctor working is so absurd it would be rejected immediately if not because chatgpt hype.

The teleprompter bit makes no sense. Do you feel less confident in your doctor after learning that any kind of serious treatment usually involves a checklist ? Following a script isn't a sign of lacking skills - it's a sign of being wise enough to recognize human limitations. Medical care isn't improv comedy.

You’re assuming it’s used as a teleprompter. More likely it’s a conversation guide.

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#214

Post by Calavar »

ClumsyPilot wrote: post #207
Calavar wrote: post #200

Earlier quoted context omitted.

For what it's worth, I'm a doctor and I find this story hard to believe. It does sound to me like wishful thinking by an AI hype guy. In my experience, these sorts of situations happen when two things come together: 1. The patient's family has just enough medical knowledge to fall onto the wrong part of a Dunning-Kruger curve. 2. The family has certain beliefs about the medical establishment. Namely, that medical sta…

> The family has certain beliefs about the medical establishment. Namely, that medical staff (especially doctors) are trained in medical school to have an inflexible way of thinking This belief saved my mother's life. We came with complainst of lung issies, for some reason doctors decided allergies are mosh likely but we wanted an X-ray. Months later they finally did a scan and found lung cancer. I don't know if they…

Well as with most beliefs, sometimes it's wrong and sometimes it's right.

I don't know anything about your mother's case, and I don’t personally have any experience with the NHS, so I can't say if it would have been appropriate to get a CT earlier in her case. Maybe it would have been.

I do feel that we as doctors can be too rigid in our decision making sometimes, but my thoughts on this are quite nuanced and probably very different from the average layperson's. It's probably also a multi-page writeup, so I won't get into it here.

In general though, CT scans cause cancer in about 1 out of every 1000 cases [1], so if we ordered them on everyone who asked, we could very well cause more cancer than we diagnosed.

In the US, we already have system that is much more patient driven and much more aggressive on testing than most of western Europe, and we still have worse health outcomes.

I agree with your concerns about Chat GPT. Patients want face to face time and I think they deserve that. Unfortunately that's low hanging fruit for C-suite executives to pick at when cutting costs.

[1] https://www.bmj.com/content/346/bmj.f2360

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#215

Post by catchnear4321 »

bemusedthrow75 wrote: post #196
catchnear4321 wrote: post #194

Earlier quoted context omitted.

> The bit about hiring someone with empathy was quite clearly in my second point, which was aimed at the HN audience, not at the doctor. then the assumption of ability to hire or impact hiring is broader. the article author is quite biased, yes. part of the problem with current “ai” - bias from all sides. chatgpt is not magical. it is a tool. a novel one. that many wish to sell. none of that lessens the possibilities…

> then the assumption of ability to hire or impact hiring is broader. It is, but so what? What I am suggesting is that using AI tools to fake empathy is worse than not being able to do empathy yourself. This should be evident. If you have a task that requires significant empathy (a patient advocate, a customer advocate, a serious complaints line, anything with extensive exposure to the emotions of your customer) and…

> What I am suggesting is that using AI tools to fake empathy is worse than not being able to do empathy yourself. This should be evident.

you miss the grain of truth in the advertisement.

in a high stress situation, someone may know how to describe the necessary words but be incapable of accessing them. this is not for lack of empathy. nor are the generated words some form of fake empathy.

is the right choice to read the output like a script to an audience? maybe not. but reading it to oneself could do wonders.

there are a lot of new possibilities, as long as the tool isn’t treated as magic or abused.

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#216

Post by madeofpalk »

Applejinx wrote: post #189

I'd bet money that the Medium article has also been, in part, rewritten by ChatGPT. Here's why: the story being told is that of a doctor, needing to communicate, being able to get through to 'regular folks' through talking down to them using ChatGPT (which is actually well suited to that task, once you acknowledge that hallucination can also mean losing stuff in translation). Then, the article itself, including the d…

Yes, the tone of voice that comes from ChatGPT reminds me of when you average a whole lot of colours and to just get gray. Or those "studies" of "here's a composite of every face in the world" that's just a pretty generic round featureless face.

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#217

Post by bemusedthrow75 »

catchnear4321 wrote: post #215
bemusedthrow75 wrote: post #196

Earlier quoted context omitted.

> then the assumption of ability to hire or impact hiring is broader. It is, but so what? What I am suggesting is that using AI tools to fake empathy is worse than not being able to do empathy yourself. This should be evident. If you have a task that requires significant empathy (a patient advocate, a customer advocate, a serious complaints line, anything with extensive exposure to the emotions of your customer) and…

> What I am suggesting is that using AI tools to fake empathy is worse than not being able to do empathy yourself. This should be evident. you miss the grain of truth in the advertisement. in a high stress situation, someone may know how to describe the necessary words but be incapable of accessing them. this is not for lack of empathy. nor are the generated words some form of fake empathy. is the right choice to rea…

I missed no grain of truth, IMO. But I will happily stipulate it is just an advertisement.

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#218

Post by easyThrowaway »

Xen9 wrote: post #192
Applejinx wrote: post #189

I'd bet money that the Medium article has also been, in part, rewritten by ChatGPT. Here's why: the story being told is that of a doctor, needing to communicate, being able to get through to 'regular folks' through talking down to them using ChatGPT (which is actually well suited to that task, once you acknowledge that hallucination can also mean losing stuff in translation). Then, the article itself, including the d…

"...hallucinatory happytalk that paints with a very broad brush, and delivers not the tone you want, but an abstracted 'popularized' tone likely to convince large numbers of stupid and gullible people." You capture essence beautifully.

A quick way I've found to recognize ChatGPT-written text is upon reading it aloud, it 's kinda obvious how constant and monotone it sounds, no matter the chosen style. It gives a very "metronomic" rhythm, almost like a krautrock song. It's kinda obvious once you notice it.

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#219

Post by cced »

smoldesu wrote: post #42
cced wrote: post #34

Earlier quoted context omitted.

How much time does this take to get produced locally?

I run a similar Vicuna bot on a free ARM VPS from Oracle. Inference takes ~20 seconds for ~200 tokens on the CPU, so it should stream results about half as fast as ChatGPT. ... however , that was on cheapo ARM hardware. On a hospital budget I bet you could beat OpenAI with a local model pre-mapped in GPU memory on a 3090.

Do you guys have any reference material on how to get this started?

Re: I’m an ER doctor. Here’s how I’m already using ChatGPT to help treat patients

#220

Post by cced »

thatcherc wrote: post #86
cced wrote: post #34

Earlier quoted context omitted.

How much time does this take to get produced locally?

Not OP, and I'd have to see the prompt to run it on my set up, but I but together a refurbished HP Chromebox with some extra RAM and an SSD for a total of $250 and it'll run all the 13B models about as fast as I can type on my phone, just as a reference point. I'm consistently impressed by it even if it's not blazing fast.

Do you guys have any reference material on how to get this started?
Post reply on HN