引用tannhaeuser · 2024-10
“has been also explored recently in one form or another in academia eg [3] also referenced elsewhere in this thread.”
指出学界也已探索过让 LLM 调用外部工具做规划,并给出这篇作参考
主张自回归 LLM 不能自行规划,需与外部验证器组成 LLM-Modulo
这是一篇立场论文:作者认为「LLM 只要提示得当就能规划推理」和「LLM 只配做格式翻译再交给符号求解器」两种极端都不对。他们主张自回归 LLM 自身不能做规划或自我验证,应被视为通用的近似知识源,并提出 LLM-Modulo 框架,让 LLM 与外部基于模型的验证器进行更紧密的双向交互,验证器所用的模型本身也可借助 LLM 获得。
评论者引用它说明自回归模型无法单独完成自我验证和长程规划,应当当作高召回的想法生成器,再由可靠的外部验证器把关。
“has been also explored recently in one form or another in academia eg [3] also referenced elsewhere in this thread.”
“position paper argues that auto-regressive models simply can't do self-verification or long-horizon planning on their own and should instead be treated as high-recall idea generators wrapped by a single, sound verifier [3].”
“very much in the spirit of the LLM-Modulo framework”