3 ms·
I would think an LLM trained on an accurate description of virtually every medical diagnosis ever made (along with a description of the treatment and outcome) w
by wizofaus 4y ago
I would think an LLM trained on an accurate description of virtually every medical diagnosis ever made (along with a description of the treatment and outcome) would be far more capable of diagnosing almost any newly described condition better than any human doctor. What I'd expect is a good deal further off is a AI- controlled surgical robots, but we'll likely see them used within our lifetimes.
- WA 4y agoOnly if you can provide an accurate description of the condition someone is suffering from. Without it, an LLM would probably recommend some average treatment. Basically, prompt engineering for medical conditions.
- wizofaus 4y agoAbsolutely, and clearly in many cases such an accurate description couldn't be supplied purely by the patient self-describing their symptoms. But for others it seems fairly plausible. I'd still expect to see qualified professionals to be required to actually approve the recommended treatment (including prescribing medications etc.).
- hackerlight 4y agoYou'd probably start by using the LLM as a filter, not for the final diagnosis. If the LLM thinks something is strange (say, 5% of the time) --> refer to human doctor to ensure it isn't a false positive. If the problem is a shortage of doctors, this is one solution.