Who Reasons in the Large Language Models?

Jie Shao, Jianxin Wu

Despite the impressive performance of large language models (LLMs), the process of endowing them with new capabilities--such as mathematical reasoning--remains largely empirical an…