> In order to decrease the number of samples needed from the environment, we expect that ideas related to model-based RL — where you spend a huge number of neural network flops to learn a model of the environment — will be the way to go.

Will those models be introspectible / transferrable? One thing I'm curious about is how AI's learn about novel actions / scenarios which are "fatal" in the real world? Humans generally spend a lot of time being taught these things (rather than finding out for themselves obviously) and eventually come up with a fairly good set of rules about how not to die in stupid ways.