Don’t be fooled—LLMs don’t reason

Former AlphaGo team member Thore Graepel argues that today’s language models lack the separate search and evidence-tracking mechanisms needed for trustworthy reasoning.

前 AlphaGo 团队成员托尔·格雷佩尔认为，如今的大语言模型缺少独立的搜索机制和证据追踪机制，因此难以进行可信的推理。