The future of robot learning from the perspective of large predictive models—world models and learned simulators and how they relate.

Perspectives, properties, and uses of learned simulators and how they could raise the waterline in robot learning.

Earlier thoughts on a learned simulator and how it might look, trying to extrapolate from things we were seeing with GPT-3. Some duplicate content with later posts, so I would start with those and come back to this if you want an earlier sample from my generative model.