17 ms·

> People will not know to ask about platelet count. I would. You would. > Anyway, this comment reads wildly similar to the dropbox comment. My point was that

by p_ing 2y ago

> People will not know to ask about platelet count. I would. You would. > Anyway, this comment reads wildly similar to the dropbox comment. My point was that this isn't revolutionary/interesting. AI w/ camera can detect apples that have bruises, but AI's cousin Vision/ML has been doing this since at least the early 2000s. Further to my point was that while yes, using all of the ways the author went about identifying the issue (picture etc) AI makes for a nicer UX, it wasn't required, nor was the invasion of privacy into the author's medical history required. The author torched a forest to figure out their issue when burning the paper mill would have done the trick.

FeepingCreature 2y ago
I would not have known that the "low platelet" on the report and the red dots were related, and I would not have brought it up or searched for it.
- bethanymarz 2y ago
  Neither did i
Ukv 2y ago
> I would. You would. I would've had no idea prior to reading this article. There seems to be real value in the LLM being able to probe with relevant diagnostic questions, then picking out "Your platelet count is alarmingly low. You should go to the ER urgently." from the three lab work panels uploaded. Even if I had properly gone through those results myself at that point (not a given, by the timing), the level of danger seems very easy to miss. > My point was that this isn't revolutionary/interesting [...] it wasn't required [...] The author torched a forest to figure out their issue when burning the paper mill would have done the trick. And you can do anything a computer can with sticks and sand. If an LLM is the best tool available (that isn't a given, but it does seem to have done well here) I don't see what makes it the equivalent to torching a forest. Its size? Energy usage? More capable than strictly necessary? Maybe factors I'd consider if designing, say, object detection for a low-powered doorbell camera - but not really things I'd worry about in what's potentially a medical emergency.

by p_ing 2y ago

FeepingCreature 2y ago

I would not have known that the "low platelet" on the report and the red dots were related, and I would not have brought it up or searched for it.

bethanymarz 2y ago
Neither did i

Ukv 2y ago

> I would. You would. I would've had no idea prior to reading this article. There seems to be real value in the LLM being able to probe with relevant diagnostic questions, then picking out "Your platelet count is alarmingly low. You should go to the ER urgently." from the three lab work panels uploaded. Even if I had properly gone through those results myself at that point (not a given, by the timing), the level of danger seems very easy to miss. > My point was that this isn't revolutionary/interesting [...] it wasn't required [...] The author torched a forest to figure out their issue when burning the paper mill would have done the trick. And you can do anything a computer can with sticks and sand. If an LLM is the best tool available (that isn't a given, but it does seem to have done well here) I don't see what makes it the equivalent to torching a forest. Its size? Energy usage? More capable than strictly necessary? Maybe factors I'd consider if designing, say, object detection for a low-powered doorbell camera - but not really things I'd worry about in what's potentially a medical emergency.