This was part of Reinforcement Learning from Offline Data and Human Feedback

On the Learning Dynamics of RLVR at the Edge of Competence

Yuejie Chi, Yale University