It's interesting watching the algorithm shred - it knows faces, but it doesn't know stuff other than faces.
This person is/is not wearing glasses.
This person is/is not wearing earrings.
This person is/is not in possession of facial hair.
This person is/is not standing in front of a plain blue background.
Kinda like the "monitors" reflected in the helmet you capped - those are kind of a reasonable reflection of a monitor wall but kind of not, and the background is what a tree would look like if Google Dream were drawing trees. His jawline is also... not great. It's the sort of thing where it works when you glance at it but doesn't hold up well to prolonged scrutiny.