Verifiable Reward
Definition
Training signals derived from automatically checkable outcomes, such as proof correctness in Lean, which allow reinforcement learning without human grading. Sanderson argues its limits include the inability to score conjecture quality.
Cast within
No topic-region cast yet — this appears once Verifiable Reward's compiled claims are aligned into a topic region's argument tree.
Claims mentioning Verifiable Reward (13)
Your notes
Loading notes...