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

221–230 of 253 posts

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

#221
> But in these scenarios, reviewing existing work is usually much faster than starting from scratch.

This is true, but it’s also a different skill than writing is. There are people who write well enough but are terrible editors. I fear for them in the generative AI workplace.

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

#222
post #219

Earlier quoted context omitted.

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?

Just one guy, me :)

But there are a lot of different routes to go. If you want to do what I've done though -

1. Sign up for Oracle Cloud and try to get their free 4 core ARM VPS (these are at-capacity very often, but free if you can get them)

2. Install the Ampere Pytorch runtime: https://cloudmarketplace.oracle.com/marketplace/en_US/adf.ta...

(can also use the ONNX or Transformers acceleration image if you need it)

3. Install your Pytorch/ONNX/Transformers software, link in system libraries and enjoy!

Whole thing works super smooth in my experience. 7B and 13B models are very usable for chatbot type applications.

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

#224

I'm having a hard time buying this anecdote. Not even the bit where its implied that multiple people read the same script multiple times and that worked for soothing people, but the actual content of the before and after; the gpt version doesn't explain it any differently and the 'compassionate' writing was absurd. My most charitable interpretation is that the anecdote is true but the article itself was written by so…

"Side note; does anyone actually think that chat gpt succeeds at rewriting things 'compassionately' or 'more intelligently'?" Often, sure. A few days ago I asked ChatGPT about a situation that needed compassion and wisdom. I thought it did way better than the 5 or 6 people I had talked to about it. https://chat.openai.com/share/2dd26347-3fbd-483f-b567-21c040... (ignore the first question, which was silly) FYI, we're…

I just read the GPT snippet in the response link you gave, and while it superficially seems well written, the text is hardly what i'd call moving or good. Instead it tries just hard enough at covering all the formulaic points of empathy and sympathy to instead emerge without a human spark of either. Essentially, it reads more like a personalized distillation of a generic committee-made sympathy note. If this is today's idea of a decent compassionate note, it's a sadly bland example of what some people consider meaningful.

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

#225
post #106

Earlier quoted context omitted.

Fake or not, people are incorporating ChatGPT into their work. We had a pipeline break today and when we looked at the PR that introduced the change it was in the PR notes that the tooling configuration was derived from a ChatGPT prompt. What do you even say or do about that?

Did ChatGPT also perform the code review?

This is a good question. There are no tests for tooling but the tooling can result in pipeline failures. Theoretically the pipeline should've failed outright when they went to merge. Instead it failed on specific conditions the developer didn't account for and weren't familiar with because the configuration wasn't something they were familiar with because they weren't building it from docs.

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

#227

Earlier quoted context omitted.

The concept is right on point though: ChatGPT is excellent at writing "fluffy" pieces full of empathy, compassion, PR talk, politician speech, and any other wordy flavor you need. As an engineer who is unable to produce such writing, this tool is quite helpful!

Reading the comments I realised that: - people are having two separate conversations: one about empathy, another about an empathetic-sounding style - many commenters use different definitions of empathy and compassion https://en.wikipedia.org/wiki/Empathy https://en.wikipedia.org/wiki/Compassion https://en.wikipedia.org/wiki/Sympathy

This touches on something, the same reason that "I'm sorry" apologies should not be written by chatbots - for some things it's most important that a human being did something, more than what they did.

When a big mistake has happened and you need to apologize, it had better be coming from you yourself. If you think you can improve it by delegating it to a fancy autocorrect software, you're missing the point of why we apologize. Or if you think a bot is more 'empathetic' than a human paying attention, you've lost touch with what it means to be empathetic. A bot can't even feel, let alone feel your situation vicariously.

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

#228
post #210

Earlier quoted context omitted.

And? What is wrong with being inundated with empathy?

It's not actual empathy. It's writing that seems empathic. The problem is that, exposed to enough writing that seems empathic but isn't, a person will learn that empathic writing is not sincere. We lose that as a society in the same way that if I check my email and see a "YOU'VE WON" subject I don't get excited, because I know it's fake. Do we really want that? A world in which when you see an empathic message, the f…

You say that as if it is known fact but I don't think that is a given at all.

The converse could be true: maybe being inundated with fake empathy is almost as good as being inundated with the real thing, ala placebo effect (or the studies which have shown that just forcing yourself to smile can actually improve your mood).

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

#229
post #148

Earlier quoted context omitted.

Looking at notes or a paper is one thing and checklists and timeouts serve an important role in safety, but being unable to hold a conversation about a fundamental concept without a script is quite another, and it is lacking a skill by definition.

Really depends on the situation. I have no doubts a typical doctor could easily ELI5[0] you anything about the procedures or treatments they're administering - once well fed, well rested, and somewhat relaxed. But if you're catching them as they're overworked and asking them to context-switch on the spot, well... I'd expect it work just as well as it would with software engineers. That is, if I was subject to random…

First, it really depends on the speciality and the emphasis on the clinical practice within that specialty.

But I'll address "I'd expect it work just as well as it would with software engineers."

This is not a great comparison. Orally presenting patients to peers and explanation to non-experts is a core skill of medical training and fundamental to clinical practice. Understanding the essential fundamentals of a handful of extremely common conditions that can be explained in simple terms is something that the majority of doctors will have to do thousands of times throughout residency possibly at the tail end of a 24. [I'm actually not in favor of it - because it often borders or is hazing with little verified educational value - but a historically common and still present practice in some places is making the intern present critically ill patients to the day team and attending after they are coming off of a 24 hour shift without any notes. Speaking of context switches, they will often have 5 or more patients like this. Even if not this extreme, the point is, doctor and software engineer training have little overlap (eg why numerous fools trying to make the next stack overflow for MDs have never succeeded).

As for this example, I can't overemphasize how common pulmonary edema and volume overload are presenting findings in the ED. This is like an experienced programmer going to ChatGPT to explain the addition operator in Javascript, which you could do, but would it be unfair to expect someone to come up with an explanation on the spot? Maybe, but then probably medicine is not a career for you. It does emphasize a different set of skill sets. And yes picked that example on purpose, because maybe one doesn't remember all the stupid implicit type conversion rules but one can still come up with a basic explanation.

As a doctor with a mixed socio-economic patient population if I just say "your family member has pulmonary edema" I'm already getting a fucking blank stare much of the time, especially if it is new. I might as well just tell them they have dropsy. I can even tailor patient education to the audience as well after 10 years of doing this for a living, maybe someone wants an ELI6 (it is often the tech people, some of them are alright - though that crowds tends to often want the iamverysmart explanation - or like many on this site just explain why I'm an idiot that knows their field less than they do).

As for specialty, people go into pathology and radiology for instance to avoid all this once done with medical school, but that is only a portion of doctors. ELI5 is relative too - consulting subspecialists and radiologists must ELI5 to their more generalist colleagues.

> asking them to context-switch on the spot

I mean this is part and parcel of hospital medicine. 5 years out of residency one can easily run through a list and handoff 20 patients with a few notes on a single sheet of paper, whereas a medical student will often have an awkward folding clipboard with a ream of notes for their 3 patients.

> If I was subject to random ELI5 requests during a busy work period, you'd bet I'd start preparing notes up front

I need a few notes to remind me of a patient's clinical status in reference to their core problems. I absolutely do not need notes on how to converse with patient's which are 99% of the time the same things over and over. Despite what's on TV, medicine is overwhelmingly routine - the drama is usually the social issues.

> and probably put them on a Wiki, and then give the people asking me a link to that wiki, and politely tell them to RTFM and GTFO

Yeah, the starting point of this approach for the typical doctor and the typical techie are unsurprisingly starkly different. Telling patients to "RTFM" isn't particular winning -- and there is a whole field of academic study simply related to Patient Education.

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

#230

Earlier quoted context omitted.

"Side note; does anyone actually think that chat gpt succeeds at rewriting things 'compassionately' or 'more intelligently'?" Often, sure. A few days ago I asked ChatGPT about a situation that needed compassion and wisdom. I thought it did way better than the 5 or 6 people I had talked to about it. https://chat.openai.com/share/2dd26347-3fbd-483f-b567-21c040... (ignore the first question, which was silly) FYI, we're…

I just read the GPT snippet in the response link you gave, and while it superficially seems well written, the text is hardly what i'd call moving or good. Instead it tries just hard enough at covering all the formulaic points of empathy and sympathy to instead emerge without a human spark of either. Essentially, it reads more like a personalized distillation of a generic committee-made sympathy note. If this is today…

Well, I guess the difference it that it didn't write it for you, it wrote it for me. And it was exactly what I needed to hear. And exactly not what humans that I talked about the situation to were providing, who tended to be all about "here's how I fixed the (supposedly) similar situation with my own dog". (which is unhelpful, given that I've had other dogs and this one is very, very different) For whatever reason, I guess they didn't think it was important to say that it is a heartbreaking situation, they didn't say "It's a tough situation, but your compassion and consideration for your dog's well-being is commendable." They didn't say "As much as we'd like to, we can't change the innate behaviors and instincts of certain breeds." They just skipped that part, like it didn't need to be said.

But that's what I needed to hear. I didn't need whatever "human spark" that you imagine (but that I personally doubt you can actually demonstrate).

So on that note, why don't you show me what you'd write, if you were trying to be helpful to someone in that situation? I'm genuinely interested in seeing if you can write something that isn't formulaic, that has that human spark you speak of, and that is otherwise better than the response that a machine gave me.

I get it if that's too much time and effort to do, but really, it shouldn't take a lot of time at all. At least, not if you are suggesting that a doctor in an ER should take a break from saving lives to do such a thing. So please, just write a paragraph or two that shows what you'd say if a friend came to you looking for empathy in such a situation.

"it's a sadly bland example of what some people consider meaningful."

Again, let's hear what you think is better, the ball is in your court. As it stands, based solely on my interaction with you as well as my above interaction with ChatGPT, it looks like machines do empathy better than humans. But I am open minded to the idea that you can actually do better than the machine. Let's hear it.

Post reply on HN