Instead they solved it much more elegantly by just live-processing the video to re-aim your pupils. They’ve been doing it for years and most people don’t even know it which is how you know they did a great job. I’ve never noticed it looking creepy or artifact-y.

I think this is great, but doesn't it hinder the ability to naturally not look in the eyes? Or were they just flaws seen in the first demos? I haven't used this kind of system.

In any case, a camera behind the screen doesn't completely solve the problem anyway, because the screen can have multiple people. For this application you need either a camera per person, or the pupil re-aiming postprocessing, to give the illusion of presence.

Samsung already sells phones where the selfie camera is under the screen. Since the screens are not (and cannot be) fully transparent, the image quality suffers.