"Every single one" is overstating the case. For one example, U-net based image segmentation methods do better than humans at identifying anomalies on scans.
Replies
LLM hallucinations are a different kind of wrongness from simple statistical error, because transformer-based deep recurrent networks do a much more complex thing than, say, an image classifier.
aug 26, 2025, 11:13 pm • 0 0 • view
I am not saying AI-based facial recognition systems should be used the way that they are, but your argument against them does need to be more nuanced.
aug 26, 2025, 11:13 pm • 0 0 • view
No it doesnt. What needs to be more nuanced is peoples understanding of the goddamn biology. This is not a simple task and peoples lives are on the line. There are plenty of examples of these things fucking up just from one machines images to another because of contrast values. I dont trust it.
aug 27, 2025, 12:15 am • 0 0 • view