Prior as Bootstrap for RL

Definition

The idea that a reasonable prior over human data is a necessary kickstart for reinforcement learning to reach high-level performance, since models cannot be RLed to feats like IMO Gold or full application coding purely from scratch.

Cast within

No topic-region cast yet — this appears once Prior as Bootstrap for RL's compiled claims are aligned into a topic region's argument tree.

Claims mentioning Prior as Bootstrap for RL (1)

Your notes

Loading notes...