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

I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

inflecthealth.medium.com

Post reply on HN
Print view View original

71–80 of 368 posts

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#71

Post by dekhn »

mabbo wrote: post #39

It got 50% right with a general model. That same model can do a pretty good job at a thousand other skills. Imagine what it can do if it's trained for exactly this purpose.

https://blog.google/technology/health/ai-llm-medpalm-researc... is an example of a more tuned model

At some point people are going to notice that the best of these systems disagree with the experts who disagree with each other on the harder problems. That's one of the issues with AlphaFold- the accuracy of their predictions are so good, its predictions have metric values as good as the limits of the experimental technology used to create the golden labels.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#72

Post by Enginerrrd »

jcims wrote: post #48

I know everyone scoffs at the concept of 'prompt engineer', but it really is an essential craft that we're going to have to come to terms with when interacting with large language models. Seeking suggestions on a more comprehensive prompt: https://sharegpt.com/c/sckAPvV Trying it out: https://sharegpt.com/c/LbpEIxi

I agree. I'm a civil engineer / project manager and so far I've been VERY impressed with chatGPT and, in particular GPT-4.

However, a huge part of my job has always been translating vague desires into very precise specifications with constraints and expectations. Going further, it has often been my job to take those specs/constraints and then break them into chunks and feed them to junior staff who are often very smart, but lack domain specific context and knowledge. Giving them a bad prompt produces bad results.

This article seems to be based largely on data collected with a rather poorly engineered prompt, IMO. He asked it a question that would be reasonable to ask a fellow physician. The problem is GPT is NOT a fellow physician with domain specific context and knowledge, and isn't aware of a bunch of implicit expectations they didn't realize they had. However, I actually think there's a really good chance that a better worded prompt would have scored a lot better here.

This type of communication skill has always been hard for a lot of people, and will remain in high demand for a long time.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#73

Post by paradite »

colonCapitalDee wrote: post #53
bentcorner wrote: post #28

Earlier quoted context omitted.

> ChatGPT does not ask for clarification, because it is not a knowledge base trying to find an answer. All it does is figure out what character is statistically most likely to come next, it has no heuristic to know that there is a task it hasn't fully completed. It absolutely can, but it requires the right prompts. If you just say "give me a diagnosis" it won't, but if you prompt it with a "you are a doctor and I am…

ChatGPT's heuristic may lead it to emit a sequence of words that suggest to English speakers that they should provide clarification, but ChatGPT is congenitally incapable of determining that answering a question requires more information that was provided in the prompt and that clarification is needed. Rephrased, ChatGPT can ask for clarification, but only as a side effect of satisfying it's heuristic.

> but ChatGPT is congenitally incapable of determining that answering a question requires more information that was provided in the prompt and that clarification is needed.

What's stopping the ChatGPT heuristic / model from determining that? That's within the capabilities of a statistical model.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#74

Post by sixothree »

yieldcrv wrote: post #58
sixothree wrote: post #52

Earlier quoted context omitted.

I had an opportunity today that made it useful. I was trying to find a reliable method of counting the decimal places in a double (64 bit float). I couldn't help but feeling like the responses were not quite informed and possibly dangerous. Chat GPT provided a solution, one that appeared better than most of what I had seen in the previous 15-30 minutes. I asked it twice to ensure safety and it improved its response.…

protip: ask it to write the unit test for you

Oh man. I did that by hand. Now I'm kicking myself. I tried to find non round trip numbers and such.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#75

Post by barking_biscuit »

fzzzy wrote: post #43
Maursault wrote: post #33

Earlier quoted context omitted.

> as "true" AI begins to emerge What, exactly, is that?

The definition has always been the Turing test. Nobody has been able to provide a more precise answer.

I think it's kind of silly. The real metrics we will evaluate AI progress by is economic impact first, followed by political impact.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#76

Post by gerdesj »

civilized wrote: post #37
gilbetron wrote: post #24

Being a scifi geek and AI geek and neuroscience geek for pretty much the past 40 years, I've read countless predictions and scenarios and stories about society's response as "true" AI begins to emerge. So watching it play out for real is creating this bizarre sense of deja vu combined with fascination and frustration and also some anxiety. This article and the comments in this thread are right up that alley. I mean,…

What I didn't predict is so many AI boosters being so miffed at regular folks pointing out the obvious: a half-working system may be tremendously exciting to AI enthusiasts, but it's not that useful to society. Recently we've had all these people complaining of moving goalposts, when in fact most people never had goalposts for AI in the first place, and certainly never had goalposts that remotely resembled those that…

It's a bloody calculator and a damn good one (but it can and does go mad). Use it like that and you'll be fine.

Have you actually engaged with it in any way? Its free to sign up.

By the way, what on earth is an AI booster? Do you really mean "enthusiast"?

"My mental model seems to clash with that of the boosters." Please tell us about your mental model, we can't read your mind.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#77

Post by dekhn »

fzzzy wrote: post #43
Maursault wrote: post #33

Earlier quoted context omitted.

> as "true" AI begins to emerge What, exactly, is that?

The definition has always been the Turing test. Nobody has been able to provide a more precise answer.

I think an updated version of the turing test needs to be done: combine a generative language model with a generative face model and speech model to create an interactive avatar that can converse with an individual. The individual is allowed to ask any and all questions of a series of avatars (50% of the avatars are actually humans) and judge whether they are Human or Not Human. If a particular avatar model is able to fool a representative sample population into identifying avatar models from humans at a similar rate, then it passes the test.

This still doesn't really move the needle on any of the important questions about AI, but does hasten the public perception that proving self-awareness, consciousness, or agency exists in humans without depending on subjective experience is probably impossible.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#78

Post by jacquesm »

messe wrote: post #67
jacquesm wrote: post #35

Earlier quoted context omitted.

Half the time right is worse than nothing at all.

Not when there are a lot more possible wrong answers than right answers.

That doesn't follow. 2+2 has more wrong answers than right answers, and yet, if you gave me the wrong answer to arithmetic sums half the time I'd think you were a lousy calculator. Worse still if those answers had the potential to kill me.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#79

Post by ekidd »

aflag wrote: post #51
ekidd wrote: post #15

Some of the best performances I've seen out of ChatGPT are essentially "junior programmer" level. But it still requires clear instructions and close supervision. But GPT's training data includes GitHub, and it's used to power Copilot. It has arguably been trained to be a programmer. In less familiar domains, like law or medicine, GPT has presumably undergone very limited training and tuning. It's essentially an "inte…

Not sure if I'd compare ChatGPT with a junior programmer. In my experience junior programmers tend to be builders. They will tend to code a lot of stuff and usually get reasonable results, but making some bad decisions that more experienced developers have already gone through. Inexperienced developers need supervision because otherwise they will just create heaps of code that will be hard to maintain later. ChatGPT…

> ChatGPT just doesn't do anything on its own and will never follow through with anything.

Yeah, I should be clear: the specific context in which I often see junior developer quality performance is when I'm essentially "pair programming" with it. I can say, "Write a program that does X. No, there's a closed-form solution for that, so we shouldn't write it recursively. OK, now write some tests similar to the following. You have a bug in that function."

I almost never need to actually write code—I can explain what I want and it will frequently do something plausible.

The main limit seems to be the context window. And ChatGPT's tendency to be ignore corrections and simply restate the original code.

Ironically, to get the best results, I need to call on my experience and my people skills. Someone who knew nothing about programming wouldn't be able to "mentor" the chatbot successfully.

Re: I’m a doctor: Here’s what I found when I asked ChatGPT to diagnose my patients

#80

Post by zzzeek »

jjcon wrote: post #42
zzzeek wrote: post #6

[flagged]

> because this doctor is not practicing in Texas where such a procedure might get you arrested https://texas.public.law/statutes/tex._health_and_safety_cod... Please don't spread misinformation, there is enough confusion out there already. Texas law specifically allows for the removal of ectopic pregnancies.

these specific provisions are insufficient and women are having harrowing health care experiences in Texas nonetheless due to doctors delaying or denying care out of fear of prosecution:

https://www.texastribune.org/2022/09/20/texas-abortion-ban-c...

Post reply on HN