LLMs trained with Chain-of-thought excel in reasoning capability, but often come with excessive token cost. We find that the core of reasoning capacity lies …
机构：UIUC
来源：arXiv 2609.37976 | AI4Papers 论文推荐平台