为理清大模型在数智服务中的角色定位与应用路径，基于信息加工理论梳理相关研究，从用户、服务提供方、平台及多主体交互层面归纳其应用模式。研究发现，大模型在数智服务中的应用呈现出由感知智能、认知智能、推理智能向执行智能递进演化的智能范式。其技术实施路径包括提示工程、微调、检索增强生成、强化学习等，但在意图理解、知识召回、推理幻觉和多智能体协同等方面仍面临挑战。进一步分析表明，上述挑战制约了大模型能力向数智服务效能的充分转化，大模型应用需要在信息感知、知识认知、逻辑推理与行动执行等环节形成系统性支撑。

To clarify the roles and application pathways of large language models in intelligent digital services, this study reviews relevant research through the lens of information processing theory and summarizes application patterns across users, service providers, platforms, and multi-actor interactions. The findings reveal a progressive intelligence paradigm comprising perceptual, cognitive, reasoning, and execution intelligence. Major approaches include prompt engineering, fine-tuning, retrieval-augmented generation, and reinforcement learning, while key challenges involve intent understanding, knowledge retrieval, reasoning hallucinations, and multi-agent coordination. These challenges hinder the translation of model capabilities into service effectiveness, requiring coordinated support in information perception, knowledge cognition, logical reasoning, and action execution.