Live data from Hacker News

I misused LLMs to diagnose myself and ended up bedridden for a week

blog.shortround.space

141–150 of 150 posts

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#141

Earlier quoted context omitted.

You've never had "the experience". I'll share mine. Unusual severe one sided eye pain. Go to regular doctor's, explain, get told it's a "stye" and do hot compresses. Problem gets worse, I go to urgent care. Urgent care doc takes one look at me and immediately sends me to the ER saying it's severe and she can't diagnose it because she is unqualified. Go to ER, get seen by two specialists, a general practicioner and a…

> About 70% of hospital deaths are due to preventable mistakes in the hospital. It's awful that you had a bad experience, but no. Nowhere near 70% of hospital deaths are from preventable mistakes. I would also note that in your experience, you ended up trusting a different doctor (ophthalmologist), not ChatGPT. Second opinions from other qualified professionals is a thumbs up from me.

https://www.ncbi.nlm.nih.gov/books/NBK499956/ cites >200,000 from one estimate. Harvard Health throws out 700,000 deaths per year in hospitals total. ~0.28 according to my quick math.

I generally expect such counts to be under-representative. I'm also probably conflating cases of "this person was going to die regardless but the hospital screwed something up" with "this person was not going to die but did die because the hospital screwed up." It's not clear how any source would avoid that conflation though.

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#142
post #83

Earlier quoted context omitted.

You've never had "the experience". I'll share mine. Unusual severe one sided eye pain. Go to regular doctor's, explain, get told it's a "stye" and do hot compresses. Problem gets worse, I go to urgent care. Urgent care doc takes one look at me and immediately sends me to the ER saying it's severe and she can't diagnose it because she is unqualified. Go to ER, get seen by two specialists, a general practicioner and a…

> About 70% of hospital deaths are due to preventable mistakes in the hospital. Sure sounds like an asspull stat. Extraordinary claims, extraordinary evidence. Do you have a reference for that? Care to share?

https://www.ncbi.nlm.nih.gov/books/NBK499956/

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#143

Earlier quoted context omitted.

> Getting medical info from LLMs is great and can save your life. No its not - LLMs are not medical experts. Nor are they experts for pretty much anything. They just extrapolate statistically the next token. If you fed them anti-vaxer information, theyd start recommending you to not get vaccinated so as to not obtain autism or something like that. We should not use them as experts on anything, much less so for medica…

> No its not - LLMs are not medical experts So what? That does not mean they're not very good at searching the internet and often provide useful information. > If you fed them anti-vaxer information, theyd start recommending you to not get vaccinated so as to not obtain autism or something like that. I specifically addressed this in the very short comment you're replying to, but I will repeat: >Blindly trusting medic…

What does being good in searching Internet (I'll let slide the obvious issue of hallucinating URLs) have to do with medical expertise? At all? At least with WebMD the content did not change every time you visited the same page?

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#144

Earlier quoted context omitted.

> No its not - LLMs are not medical experts So what? That does not mean they're not very good at searching the internet and often provide useful information. > If you fed them anti-vaxer information, theyd start recommending you to not get vaccinated so as to not obtain autism or something like that. I specifically addressed this in the very short comment you're replying to, but I will repeat: >Blindly trusting medic…

What does being good in searching Internet (I'll let slide the obvious issue of hallucinating URLs) have to do with medical expertise? At all? At least with WebMD the content did not change every time you visited the same page?

What does medical expertise have to do with being able to provide useful information? AI has no expertise.

> I'll let slide the obvious issue of hallucinating URLs

That indeed was an issue, but I can't remember the last time I've encountered that now that agentic web browsing is everywhere. I guess the cheapest models might still be affected by that?

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#145

Earlier quoted context omitted.

What does being good in searching Internet (I'll let slide the obvious issue of hallucinating URLs) have to do with medical expertise? At all? At least with WebMD the content did not change every time you visited the same page?

What does medical expertise have to do with being able to provide useful information? AI has no expertise. > I'll let slide the obvious issue of hallucinating URLs That indeed was an issue, but I can't remember the last time I've encountered that now that agentic web browsing is everywhere. I guess the cheapest models might still be affected by that?

Well, I guess since you have not encountered it recently, then its fine... Everyone else is just using "cheapest" models ;)

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#146

Earlier quoted context omitted.

What does medical expertise have to do with being able to provide useful information? AI has no expertise. > I'll let slide the obvious issue of hallucinating URLs That indeed was an issue, but I can't remember the last time I've encountered that now that agentic web browsing is everywhere. I guess the cheapest models might still be affected by that?

Well, I guess since you have not encountered it recently, then its fine... Everyone else is just using "cheapest" models ;)

No matter how badly the AI gets things wrong, it's fine. You're not supposed to trust it.

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#147

Earlier quoted context omitted.

Well, I guess since you have not encountered it recently, then its fine... Everyone else is just using "cheapest" models ;)

No matter how badly the AI gets things wrong, it's fine. You're not supposed to trust it.

No its bloody not fine! We were told these were "Pocked-PhDs". The investment so far could have paid for a manned mission to Mars and back! The massive advertising campaigns absolutely try to subliminally influence us into believing these are trustworthy tools, while at the same team keeping that sneaky small print about "checking the answers". Whats the goddamn point then? If I cant trust them every single time, like I can with deterministic software, and they create more work for me, because now we have to double-check the shit that they spit out, how are they actually useful, beyond trivial use cases like producing stupid brainrot videos??

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#148
post #107

Earlier quoted context omitted.

[flagged]

This is a completely unacceptable comment on HN. The guidelines make it clear we're looking for a higher standard of discourse here. These lines in particular are relevant: Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes. Comments should get more thoughtful and substantive, not less, as a topic gets more divisive. When disagreeing, please reply to the argument instead of calling nam…

Please ban me. I don't want to participate anymore.

Re: I misused LLMs to diagnose myself and ended up bedridden for a week

#149
post #107

Earlier quoted context omitted.

This is a completely unacceptable comment on HN. The guidelines make it clear we're looking for a higher standard of discourse here. These lines in particular are relevant: Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes. Comments should get more thoughtful and substantive, not less, as a topic gets more divisive. When disagreeing, please reply to the argument instead of calling nam…

Please ban me. I don't want to participate anymore.

[dead]
Post reply on HN