Depends on whether it's an ai summary of the visit or just using the ai to more accurately guess what the doctor actually said compared to traditional speech recognition (which has been in use since long before what we call modern AI).
There's definitely been some times where I've read some completely nonsensical word choices and had to sit for a minute or two trying to imagine the doctors accent and think about what sounds like those words that would make more sense in context. I feel like that's actually the kind of task an llm could really help with, especially if you can use specialty specific models, since using probabilities to guess the next word is like literally their jam. In a similar vein OCR enhancement too.
The OP sounds like more of a problem with those AI summaries they're pushing. Those are pretty easily outmatched by the doctor just putting on a diagnostic code and auto printing some premade patient ed leaflets. You really wanna up that game you can have the ai search the human made and chosen leaflet with fuzzy match and highlight any specific side effects or whatever that were mentioned in the visit but again that's still using the human made content the AI would just enhance the way it's presented.
The main problem I see is that they're trying to make the AI take over the task completely and generate it's own information when it should really just be adding better fuzzy match to searches and word prediction to scribing software.
3 Comments
RobotToaster@mander.xyz · 3 pts · 1d
I've had their human scribes do the same. I wonder what the difference in error rate is.
Apytele@sh.itjust.works · 5 pts · 1d
Depends on whether it's an ai summary of the visit or just using the ai to more accurately guess what the doctor actually said compared to traditional speech recognition (which has been in use since long before what we call modern AI).
There's definitely been some times where I've read some completely nonsensical word choices and had to sit for a minute or two trying to imagine the doctors accent and think about what sounds like those words that would make more sense in context. I feel like that's actually the kind of task an llm could really help with, especially if you can use specialty specific models, since using probabilities to guess the next word is like literally their jam. In a similar vein OCR enhancement too.
The OP sounds like more of a problem with those AI summaries they're pushing. Those are pretty easily outmatched by the doctor just putting on a diagnostic code and auto printing some premade patient ed leaflets. You really wanna up that game you can have the ai search the human made and chosen leaflet with fuzzy match and highlight any specific side effects or whatever that were mentioned in the visit but again that's still using the human made content the AI would just enhance the way it's presented.
The main problem I see is that they're trying to make the AI take over the task completely and generate it's own information when it should really just be adding better fuzzy match to searches and word prediction to scribing software.
Salvo@aussie.zone · 2 pts · 1d
This bullshit needs to be regulated.