Smart labeling strategy stabilizes AI reasoning training with limited data

Original: When Self-Belief Misleads: Active Label Acquisition for Reinforcement Learning with Verifiable Rewards

Source: arxiv.org ↗