This is different.My understanding from a Tesla engineer is that the learning is ONLY done in the car itself right now. Once the neural net is complete and they turn it on to the fleet, the entire fleet will learn in unison.
Case in point: Initial Autopilot calibration where you cannot even activate the feature until it learns the basics.
Initial "learning" is basically watching you drive for some time to see how badly misaligned the camera is from install to add a corrective mask to the image before starting to analyze it.
There's no actual learning happening in car, the pre-learned Neural Net is included with the firmware and used for visual recognision of objects, that's all.