Reinforcement learning can substantially improve a reasoning teacher, but it is unclear which of those improvements survive when the teacher supervises a sma…
机构:Microsoft
来源:arXiv 2610.05974 | AI4Papers 论文推荐平台