Jeff, We know how to shape it. It’s just a LOT bigger leap. And for real time le

Jeff,
We know how to shape it.
It’s just a LOT bigger leap.
And for real time learning we need neuromorphic hardware.
Both Tesla and Boston Dymanics use world models.
World models let us test for the possibility.(consistency of external correspondence)
Now try to imagine how much more difficult it is to predict a competition between x episodes of memory of world models instead of text?
That’s the problem.
Can we struggle toward simulating AI with just text? Possibly. But that still means it’s a robot not an AI.

Reply addressees: @JeffLadish