biological priors from evolution

pretraining distribution

Both act as the start state on which RL operates, suggesting priors rather than the RL loop explain differing sample efficiency.

1 source videos