Context Bootstrapped Reinforcement Learning

Saaket Agashe, Jayanth Srinivasa, Gaowen Liu +2

Reinforcement Learning from Verifiable Rewards (RLVR) suffers from exploration inefficiency, where models struggle to generate successful rollouts, resulting in minimal learning si…