Do LLM Agents Execute the Plans They Declare? From Planning-Mode Declaration to Pattern-Specific Execution· LLM 代理执行声明的计划吗?
Large language models (LLMs) enable agents to solve long-horizon tasks by generating a plan and then executing it in an environment. However, successful planning requires two distinct capabilities: selecting an appropriate plan for the task and executing it faithfully. Existing planner--executor systems can fail at either stage, while final task success alone cannot distinguish selection from execution failures. We therefore study the Plan Declaration--Execution Gap and introduce Planning-as-Routing, where an LLM declares one of four planning modes: Predefined, Sequential, Hierarchical, or Search, and a deterministic router dispatches the task to the corresponding pattern-specific executor. Across four benchmarks and three LLMs, we find three consistent patterns. First, generic Plan+ReAct
研究 LLM 代理的计划执行差距,提出 Planning-as-Routing 方法。
- 核心方法
- 引入 Planning-as-Routing 框架,通过 LLM 宣布四种规划模式(预定义、顺序、层次或搜索),再由确定性路由器将任务分配给特定模式的执行器。
- 适合谁读
- 研究者、工程师
- 要解决的问题
- 现有 LLM 代理系统在任务规划和执行阶段可能失败,且单凭任务最终成功难以区分是规划失败还是执行失败。
- 关键实验
- 在四个基准测试集上,使用了三种 LLM 模型进行实验,发现了三个一致的模式,但具体内容未详述。
- 主要贡献
- 系统地研究了计划声明与执行之间的差距,提出了一种新的框架来改进 LLM 代理的任务执行。
- 意义与局限
- 有助于理解 LLM 代理如何处理复杂任务,为设计更可靠的规划和执行系统提供思路。局限性在于实验仅涉及有限的模型和任务类型。