We argue that pattern recognition and step-by-step reasoning are two ends of a spectrum. A large language model (LLM) learns to reason step-by-step when data…
机构:宾夕法尼亚大学
来源:arXiv 2610.09186 | AI4Papers 论文推荐平台