"Tesla in particular has a fleet of over 100,000 cars deployed in the real world that it is constantly collecting and then annotating data from. These cars are being teleoperated by experts, making the data ideal for large-scale supervised learning."

Biological brains are extremely good continual learners. They don't need to be trained ahead of time at all for basic motor skills. They'll learn on their own and sharing information merely speeds the process up.

There is no such thing as a data problem. The scaling paradigm isn't obsolete because it doesn't work, it's obsolete because it appears to be wholly unnecessary.

Think about it, why the hell would it be necessary for every single person wanting a robot butler in the house to take a video of their kitchen and then blend it with tens of thousands of other kitchens, just so the robot knows your specific kitchen in and out? The other kitchens are irrelevant to you. You don't care about them.

I'm not affiatled with them.