Reinforcement Learning From Verifiable Rewards Rlvr On Chainofthought Reasoning

“In this paper, we develop two metrics for critically examining this assumption: Causal Importance of Reasoning (CIR)... and Sufficiency of Reasoning (SR)...”