BriefGPT.xyz
Oct, 2024
通过几何约束大语言模型导航复杂物理世界
Navigate Complex Physical Worlds via Geometrically Constrained LLM
HTML
PDF
Yongqiang Huang, Wentao Ye, Liyao Li, Junbo Zhao
TL;DR
本研究解决了使用大语言模型(LLMs)重建和构建基于文本知识的物理世界的能力缺口。研究提出了一套几何约定,并基于多层图和多智能体系统框架开发了工作流程,以提高对复杂物理世界几何和空间关系的理解。重要发现是,LLMs能够在统一的几何约定下,通过多层图实现多步骤和多目标的几何推理。
Abstract
This study investigates the potential of
Large Language Models
(LLMs) for reconstructing and constructing the physical world solely based on textual knowledge. It explores the impact of model performance on
Spatial Unde
→