The photos in the first article are self selected from Facebook and not just a plain, consistent forward facing picture. For all anyone knows that AI was selecting based on lighting or photo quality or facial expressions or simply what each participant tended to pick as their picture, likely not anything to do with the actual face at all. And he says “accurately” but it’s only 72% accurate with a training corpus of over a million data entries.
After $250,000,000 of seed funding and thousands of hours of training, our algorithm is able to accurately classify this picture of me sucking dick while burning down a police station 3 out 5 times (under optimal conditions)!
The photos in the first article are self selected from Facebook and not just a plain, consistent forward facing picture. For all anyone knows that AI was selecting based on lighting or photo quality or facial expressions or simply what each participant tended to pick as their picture, likely not anything to do with the actual face at all. And he says “accurately” but it’s only 72% accurate with a training corpus of over a million data entries.
I didn’t bother to check the other two articles.
After $250,000,000 of seed funding and thousands of hours of training, our algorithm is able to accurately classify this picture of me sucking dick while burning down a police station 3 out 5 times (under optimal conditions)!
Why would you want to do this under suboptimal conditions though? You need it to be just right to be as enjoyable as possible