|
my doctor was trying to use speech to text instead of just typing out notes and every time he got to a vaguely medical word he had to repeat it like 5 times and then give up and type it also it's great hearing him list out loud poo poo that's wrong with me in front of me but not to me
|
# ? Jul 24, 2023 03:26 |
|
|
# ? Jun 9, 2024 06:11 |
|
training a transcription model on medical transcriptions would probably be quite doable if they actually tried to do that though the existing ones presumably just aren't trained on that kind of data
|
# ? Jul 24, 2023 03:33 |
|
lol my doctor was just complaining to me about how garbage the medical speech to text software he uses is and then asked if i could fix it
|
# ? Jul 24, 2023 03:38 |
|
mystes posted:training a transcription model on medical transcriptions would probably be quite doable if they actually tried to do that though i imagine it's not the training data that's the problem, it's the speech to text. theres a lot of jargon and pronunciation can be all over the place i mean, that's not going to do anything for the llm spitting out hallucinatory data or stop it from randomly switching medicine names around or something, but basic summarization in a litany of formats is not a problem for them until they start making things up
|
# ? Jul 24, 2023 03:40 |
|
Beeftweeter posted:i imagine it's not the training data that's the problem, it's the speech to text. theres a lot of jargon and pronunciation can be all over the place Of course if they're just using chatgpt there's no training data involved in the summarization process so they can't fix that part if they wanted to
|
# ? Jul 24, 2023 03:42 |
|
mystes posted:training a transcription model on medical transcriptions would probably be quite doable if they actually tried to do that though apparently according to him on the screen where you enter prescriptions if you say weird drug names it gets it every time, but if you say them anywhere else it has no idea what you're saying
|
# ? Jul 24, 2023 03:42 |
|
medical software is enterprise software for all the ancient enterprise software reasons so you can never expect it to be good and if its good you can expect it to rapidly go to poo poo they still havent gotten off of mumps, why would you expect them to be able to do anything in any way
|
# ? Jul 24, 2023 03:44 |
|
there are a handful of places where this kind of AI poo poo might be useful in healthcare: suggestions to the doctor based on medical history (dont actually need AI for this, just codified policy and procedure) suggestions to the doctor during the consult based on speech to text keyword triggers that still require the doctor to accept or ignore (also probably dont need AI for this, but if trained properly it could get good at speech to text) mass analysis of transcripts and medical data/history/outcomes to find unknown correlations that can be passed off to a human for study.
|
# ? Jul 24, 2023 03:44 |
|
medical / AI related, this 'long but worth it' Tumblr post about the KOKOBOT mental health bit that Tumble management have partnered with has some yikes moments: https://www.tumblr.com/charlesfosterofdensen/723689352281178112 If you're having a type of issue, it will connect you to another user with a similar issue to try and work things out between you, I don't have to explain how bad an idea this is even surface-level before the opportunity for abuse of the system. edit: ugh the Tumblr management are trying to ride the twitter discontent wave to gee up account signups by adding a 'log in to read more' blocker, that's only just been added. fucks sake. NoneMoreNegative fucked around with this message at 03:48 on Jul 24, 2023 |
# ? Jul 24, 2023 03:44 |
|
NoneMoreNegative posted:medical / AI related, this 'long but worth it' Tumblr post about the KOKOBOT mental health bit that Tumble management have partnered with has some yikes moments: But of course once you add people trying to abuse it jesus christ like just the thought of someone telling it "oh yeah I'm totally an abused minor, please connect me to some abused minors thanks" is just loving terrifying
|
# ? Jul 24, 2023 03:47 |
|
about 15-35% of all doctor time throughout the united states is spent dealing with this bullshit, with software production value i would not accept from high school interns. direct traceable fatalities in the hundreds, costs in the tens to hundreds of billions. they need to learn to write a db crud app before trying ml poo poo
|
# ? Jul 24, 2023 03:50 |
|
mystes posted:This is crazy because even assuming it actually works as intended, why would connecting people to other people with the same issues even be a good idea? peer support without mediation is a great idea, just look at infernal machines fucked around with this message at 03:56 on Jul 24, 2023 |
# ? Jul 24, 2023 03:51 |
|
mystes posted:This is crazy because even assuming it actually works as intended, why would connecting people to other people with the same issues even be a good idea? I wish you could read the whole thing without a signup bug in the way
|
# ? Jul 24, 2023 03:52 |
|
NoneMoreNegative posted:I wish you could read the whole thing without a signup bug in the way if you have ublock origin or a similar rules based filter, this should solve the login wall problem code:
|
# ? Jul 24, 2023 03:55 |
|
yeah that's horrifying and it pretty much seems like the worst idea ever. It really seems like someone should stop it but I guess it's not illegal or anything somehow so it's probably going to take a lawsuit after children are harmed by it?
|
# ? Jul 24, 2023 03:55 |
|
NoneMoreNegative posted:
that's insane
|
# ? Jul 24, 2023 03:56 |
|
bob dobbs is dead posted:about 15-35% of all doctor time throughout the united states is spent dealing with this bullshit, with software production value i would not accept from high school interns. direct traceable fatalities in the hundreds, costs in the tens to hundreds of billions. they need to learn to write a db crud app before trying ml poo poo one of the biggest problems with EHRs is that they're designed to be customized for every provider that uses them which will always create bad outcomes, not just in terms of software quality but also in the poor medical processes they promote. that said if traceable deaths to EHRs is only in the hundreds its probably doing waaaaay better than written paperwork.
|
# ? Jul 24, 2023 03:56 |
|
and yeah that's a lawsuit waiting to happen, goddamn
|
# ? Jul 24, 2023 03:56 |
|
last page, but jeff issues outstanding: setting it so that people have to be logged in to buy other people's avatars. this would give admins the power to find and permaban people who buy others really loving lovely abusive avatars. this is solely a jeff thing that everyone wants done but he hasn't done yet. there is not a technical limitation to this. it is literally just in jeff's hands. the front page. jeff promised to revamp this and has failed to do absolutely anything with it. you have to work to get to the front page now, as somethingawful.com just redirects directly to the forum. jeff has also failed to give us any of the updates re: forums financing/operating costs/other stats. these are just three things while there are further policy and tech things that he could be working on, but he doesn't. he's actively disengaged, and i wish he'd hire a community manager so that poo poo could get done.
|
# ? Jul 24, 2023 03:56 |
|
Shaggar posted:one of the biggest problems with EHRs is that they're designed to be customized for every provider that uses them which will always create bad outcomes, not just in terms of software quality but also in the poor medical processes they promote. i do some support work for a consulting radiologist and every single time she gets patient record data from a hospital, it's in a different format and a different app. even the metadata for the images is formatted differently depending on what brand of machine it went through, and half the time the basic poo poo like time and date are wrong
|
# ? Jul 24, 2023 03:58 |
|
infernal machines posted:peer support without mediation is a great idea, just look at or like, most of reddit
|
# ? Jul 24, 2023 04:01 |
|
NoneMoreNegative posted:
yo what the gently caress, who would ever think this is a good idea to make
|
# ? Jul 24, 2023 04:02 |
|
Midjack posted:yo what the gently caress, who would ever think this is a good idea to make people selling llms as a solution to anything this is the "exuberance" phase of new tech and there are plenty of charlatans willing to tell you it will cure what ails you and everyone else too
|
# ? Jul 24, 2023 04:03 |
|
Mr. Nice! posted:last page, but jeff issues outstanding: there's a front page?
|
# ? Jul 24, 2023 04:03 |
|
rotor posted:yeah this agreed, unless it’s really funny like doublesur
|
# ? Jul 24, 2023 04:03 |
|
nudgenudgetilt posted:there's a front page? not anymore there isn't
|
# ? Jul 24, 2023 04:04 |
|
fart simpson posted:agreed, unless it’s really funny like doublesur
|
# ? Jul 24, 2023 04:04 |
|
not much, what's up with you?
|
# ? Jul 24, 2023 04:05 |
|
NoneMoreNegative posted:
this system was designed by someone who has never had any actual problems ever and it shows
|
# ? Jul 24, 2023 04:06 |
|
Shame Boy posted:or like, most of reddit absolutely damning
|
# ? Jul 24, 2023 04:06 |
|
infernal machines posted:i do some support work for a consulting radiologist and every single time she gets patient record data from a hospital, it's in a different format and a different app. even the metadata for the images is formatted differently depending on what brand of machine it went through, and half the time the basic poo poo like time and date are wrong In theory CMS was supposed to come up with standards for all EHRs to use and then give money to providers to buy EHRs that meet those standards but all CMS actually knows how to do is the "give money to providers" part so the standards are a joke.
|
# ? Jul 24, 2023 04:06 |
|
Shame Boy posted:this system was designed by someone who has never had any actual problems ever and it shows
|
# ? Jul 24, 2023 04:10 |
|
mystes posted:It was probably designed entirely by people who were popular and or bullies in high school no mystes, it's a piece of software
|
# ? Jul 24, 2023 04:11 |
|
one interesting avenue of research on ai in medicine is some degree of guidance for doctors on treatment plans. not like where you let the ai doctor diagnose you and completely design your path through the healthcare system, but there are a lot of cases where somebody might be suffering multiple morbidities. and many times the “optimal” treatment plan for one condition will not only be contraindicated for another but also change how a third, fourth, or fifth might react. that’s a lot of information to keep straight in the moment post hole digger posted:it was snape kills dumbledore whenever that book came out too once lol as i die my last thought will probably be that dude driving up to a barnes and noble and screaming that out with a woman running after him, going “you biiitch”
|
# ? Jul 24, 2023 04:11 |
|
Shaggar posted:In theory CMS was supposed to come up with standards for all EHRs to use and then give money to providers to buy EHRs that meet those standards but all CMS actually knows how to do is the "give money to providers" part so the standards are a joke. the hospital I worked at had a small dev team that just handled HL7 translations between systems in just that one hospital because nothing followed the standard at all I can’t loving imagine doing that as a job 40 hours a week.
|
# ? Jul 24, 2023 04:30 |
|
idk about the whole medical ai thing. on one hand my doctor has given me provably bad advice that makes sense at face level. on the other i'm kind of cooling on any enthusiasm i had for ai poo poo. it's often embarrassingly wrong, unwilling to admit that it's wrong, and is just kind of incapable of nuance. both sides suck in a lot of ways. the humans will be happy to bring their biases to the table and gently caress up care for everyone. the machines just follow the biases of all the data they have and can only just spit out what statistically goes together which is its own problem.
|
# ? Jul 24, 2023 04:45 |
|
for the record i only mean academic research, i’ve yet to see a private company do something really great with it in a patient-facing sense
|
# ? Jul 24, 2023 04:47 |
|
Eeyo posted:it's often embarrassingly wrong, unwilling to admit that it's wrong, and is just kind of incapable of nuance. okay, sure, but it must be worse than your average doctor somehow
|
# ? Jul 24, 2023 04:49 |
|
yeah that does also describes lots of doctors. it is hosed either way, but i guess i'd just prefer to take my odds with the humans. at least i can read a human, those loving machines are black boxes.
|
# ? Jul 24, 2023 04:54 |
|
|
# ? Jun 9, 2024 06:11 |
|
at least i can sue a doctor for malpractice
|
# ? Jul 24, 2023 04:57 |