So if a model with exactly the same architecture controls a robot then it's suddenly sentient or what? "They generate actions in the real world"