this post was submitted on 20 Dec 2024
637 points (99.4% liked)
196
16747 readers
1975 users here now
Be sure to follow the rule before you head out.
Rule: You must post before you leave.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
they do not and saying this sort of thing actually harms people. do not trust an llm with anything medical, ever.
llms have no conception (of anything) of truth, you are getting a probabilistic bullshit. in a literal sense.
You do not trust them. Fact check their recommendations. Check the diagnose and see if the symptoms match on google. And if it is something possibly grave then go to a doctor.
But many people do not feel like going to the doctor for every small thing they wonder about.
the effort to fact check the bussshrt machine is exactly the same hs not using the bullshit machine inethe first place.
No it helps suggest the same answer faster.
One could argue who needs Google when you have books as well.
except the "answer" is likely to be wrong such that the same search needs to be done sans ai to verify it. you do not understand ai if you think they provide the same answers, they are probabilistic bullshit machines. Google does not provide answers (well didnt used to, when it was better). it catalogued places to find them. your analogy is more accurately "who needs libraries when you have books".
Do not ever use llm for medical advice, doing so will literally harm and kill people. if you want bullshit, just make some up, you dont even need the llm for that.
i will repeat this in no uncertain terms, llm cannot provide information, trusting one with your healthcare in any capacity is Stupid. it is stupid, harmful, and will kill people. if you value your health at all, just use normal web searches to get info from medical websites, call nurse hotlines, use urgent cares, call doctors offices for advice or appointments.
My dude you don't need to believe all medical advice it gives. Just use it to get a feel in the right direction and then check whether the symptoms match. It will often suggest multiple options and rate them by likelyhood.
I do not think you understand how much training these LLM's have on medical material. They can accurately diagnose almost any common disease it is not like WebMD always suggesting you have stage 5 cancer.
Seriously try it once before going full anti AI mode.
You quite literally cannot trust them, their produced information entropy is too high. I understand how much training they have on medical text, you dont understand how little that means. These models are fundamentally incapable of assessing the truth of a statement, you are using something you dont even understand to give you advice about something it cannot reliably give and lack the expertise needed to understand how accurate they actually are at any given answer, on a topic that directly influences your actual physical wellbeing!
"just try it bro it's good i promise" you should actually prompt an llm about a topic you know about in detail. the amount of errors are rampant, then apply that same inaccuracy to topics you know nothing about.
my next recommendation is that since you are not a healthcare professional, do not give medical advice like "use llm" as you personally clearly cannot verify the accuracy of llm for this role.
If you want to visit a Doctor for every minor thing feel welcome. So far LLM's have correctly predicted every health issue I have had and provided better and more accurate information than the doctor visit afterwards.
This does not mean they are infaillible. But you can easily check what they suggest and see whether the symptoms match other websites and the doctors description.
lol no
"easily just do the same search you would have after" truly the llm is very helpful and not just an uncertainty adding middle step that through your own admission you rely on over medical professionals.
Do you believe doctors to be all wise? They are similarly prone to error. Their head is not Wikipedia.
Doctors have liability and the ability to self regulate their confidence and understand certainty. Also lol "similarly prone to error" no, human cognition is not a transformer.
Do not give medical advice. Neither you nor any llm are licensed and capable of doing so. Yes, that means you should be held legally liable if that advice ever leads to harm, as should ai companies for convincing you and others of their grift.