Method makes AI math reasoning training more efficient by fixing wasted learning

Original: CATPO: Critique-Augmented Tree Policy Optimization

Source: arxiv.org ↗