Core Concepts
LLMsを計画に効果的に組み込む方法を研究し、問題解決能力と効率性を向上させる。
Abstract
大規模言語モデル(LLMs)を使用した新しい計画フレームワークが提案され、様々なドメインでその有効性が実証されました。この研究は、LLMsを深く埋め込んでオフシェルフの計画フレームワークに組み合わせることで、問題解決成功率や探索効率を著しく向上させる可能性があることを示しています。また、LLMsのプロンプティングやソートなど特定の手法が問題解決に与える影響も詳細に分析されています。
Stats
Plan synthesis aims to generate a course of actions or policies to transit given initial states to goal states.
LLMs are evaluated to be quite ineffective in planning autonomously.
LLMs can provide helpful information for planning but cannot solve planning problems solely.
The proposed LLMs4Plan approach significantly reduces the total number of expansion actions and mutually exclusive actions compared to traditional graph-based planning.
The success rate of problem resolution is improved with the integration of LLMs into the planning framework.
Quotes
"Inspired by the result of loosely using plans generated by LLMs as seed plans, we are curious if it is possible to “dig” more helpful information from LLMs to assist planning deeply."
"Through this study, we provide new clues for how to deeply embed LLMs into off-the-shelf planning frameworks."
"Both pruning and sorting contribute to enhanced search efficiency, with their combination amplifying this effect."