Comment by thepotatodude
2 days ago
We would still have this problem. The heavier models make mistakes at too high a rate vs. a physician. Especially on imaging data. Real world data and patient presentations often deviate from the textbooks they are trained on.
-Med student
That's a different class of problem. It will do just fine on text based queries spanning a few pages. Probably better than the average physician (average over all countries).
I do agree that LLM's are not there yet in the image part.