Google AI Unveils Supervised Reinforcement Learning (SRL): A Step Wise Framework with Expert Trajectories to Teach Small Language Models to Reason through Hard Problems