Synthetic task variants match human-written ones for AI reinforcement learning

Original: Trading Human Curation for Synthetic Augmentation in RLVR

Source: arxiv.org ↗