Aeye man bad! It's so dumb. You can see right in the video. They trained the AI to learn what crude readings from the radio corresponded to what positions of people in the room. The precise 3D models of people it output were not things it actually determined from inputs. They are very fuzzy guesses, just showing the precise human models that were fed to it with the visible-light camera. This is a big problem with how AI output is presented, it's precise and coherent even though it's full of uncertainty.