The technique, called Reinforcement Learning with Verifiable Rewards with Self-Distillation (RLSD), combines the reliable ...