Anyway, this comment reads wildly similar to the dropbox comment.
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.
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.
I agree with you about the Dropbox comment though. This whole thing is about having a polished user experience rather than bouncing between Google searches, Google Lens, and various other websites.
I think that should then be enough to know to Google "Low platelet count". The person may not know the significance of that immediately, true.
Dropbox?
https://news.ycombinator.com/item?id=9224
Basically, something along the lines of (paraphrasing) 'this can be done today by using existing tooling'
- Accept cookies?
- Turn off your ad blocker please!
- Why not sign up for our newsletter?
- Here's an interstitial ad
- Like us on social media!
...oh by the way, you should probably go to the ER.
Btw I agree AI can help, my issue is with greedy people salivating for money while attempting to do "AI doctor", "AI teacher" "AI physiatrist" , this important AIs should be build by non profits and correctly checked by competent people for safety. Because of $$ reasons this AIs will be marketed with more abilities then they have and the downsides will be hidden, again because of $$