New Training Trick Helps AI Explore More Answers and Boost Math Skills

Original: Instruction-Conditioned Exploration with Asymmetric Reinforcement Learning and Self-Distillation

Source: arxiv.org ↗