[sosodev]

Isn't that assuming training methods remain the same?

It seems like learning from the environment will be a requirement for robots to scale. My understanding is that research has been yielding new architectures that might have that type of real-time, general intelligence but we haven't seen that similarly large investment yet.