I don’t know how it works with radiology, but my experience in software engineering is that reviewing the slop code takes more time than writing it, especially when it is crap and you have to send it back again and again and then review it.
At this point I either have to go through honestly which is extremely slow and frustrating to both sides, or accept the slop without review and then deal with tech debt later.
Well in radiology we are searching images for specific findings so the generative slop problem isn’t the issue for us, it will be being overwhelmed with false positives or false negatives with a time pressure to go faster. I’ve been trying to follow the impacts of these models on the coding professions and I do not envy you at all. It really does seem like a rock and a hard place right now.
I don’t know how it works with radiology, but my experience in software engineering is that reviewing the slop code takes more time than writing it, especially when it is crap and you have to send it back again and again and then review it.
At this point I either have to go through honestly which is extremely slow and frustrating to both sides, or accept the slop without review and then deal with tech debt later.
Both options are bad.
Well in radiology we are searching images for specific findings so the generative slop problem isn’t the issue for us, it will be being overwhelmed with false positives or false negatives with a time pressure to go faster. I’ve been trying to follow the impacts of these models on the coding professions and I do not envy you at all. It really does seem like a rock and a hard place right now.