Group Relative Policy Optimization

Explain group-relative advantage estimation, the policy objective, and how automatically verified kernel samples could supply the reward.