DeepSeek热潮下的双重变革:大模型的技术革新与高校图书馆服务范式的重构  

Dual Transformation Under the DeepSeek Wave:Technological Innovation in Large Models and the Restructuring of Service Paradigms in Academic Libraries

作　　者：童云海 陈建龙[1] TONG Yunhai;CHEN Jianlong

机构地区：[1]北京大学图书馆,北京100871

出　　处：《大学图书馆学报》2025年第1期66-70,共5页Journal of Academic Libraries

摘　　要：近期DeepSeek公司开源发布了一系列具有自主知识产权的大模型,成为全球人工智能领域的焦点。本文解读了DeepSeek的核心技术和工程层面的创新点,剖析了技术创新在支撑计算效率提升和成本控制方面的优势。以数据融合、人机协同和智能反馈为核心的新型知识服务体系为目标,探讨了大模型在高校图书馆的资源管理革新、空间功能扩展和服务模式创新等场景的落地应用的必要性和可行路径,及其对图书馆服务范式重构的推动。提出了高校图书馆在大模型支撑下的数智化转型面临的挑战和应对策略。指出智能化时代背景下,寻求技术效能与人文价值之间的平衡,是高校图书馆在智能时代面临的重要挑战和使命。Recently,DeepSeek has open-sourced a series of large-scale models with independent intellectual property rights,rapidly rising to global attention in the artificial intelligence field due to their distinct advantages of“low cost,high performance,and robust reasoning capabilities.”This advancement is poised to catalyze profound transformations in higher education.This paper deciphers DeepSeek s core technological and engineering innovations,including breakthroughs in the Mixture of Experts(MoE)architecture,Native Sparse Attention(NSA)mechanisms,and reinforcement learning optimization strategies.These innovations underpin significant improvements in computational efficiency and cost control,such as MoE s dynamic sub-model activation mechanism reducing computing power consumption,NSA enabling efficient processing of million-word-level texts,and reinforcement learning frameworks outperforming other leading models in reasoning tasks.DeepSeek s success heralds a dual transformation for academic libraries:technological innovation and restructuring of service paradigms.Guided by the principle of“demand-driven development,resource-based foundations,technology-enabled wings,and service-centric cores,”this study explores the necessity and actionable pathways for applying large models to reshape libraries.Centered on building a new knowledge service system characterized by data fusion,human-machine collaboration,and intelligent feedback,the analysis focuses on three key scenarios:resource management transformation(e.g.,intelligent cataloging of unstructured resources like lecture videos using retrieval-augmented generation),spatial service expansion(e.g.,immersive learning environments via VR/AR and distributed virtual reading spaces),and service model innovation(e.g.,AI-assisted academic writing tools and dynamic knowledge extraction platforms).These applications drive a closed-loop collaborative evolution,where technology-driven innovations and scenario-specific demands mutually reinforce iterative advancements.How

关 键 词：DeepSeek 大模型 数字化转型 服务范式 人工智能 学术伦理 

分 类 号：G250.7[文化科学—图书馆学] TP18[自动化与计算机技术—控制理论与控制工程]