摘要
大模型产业发展迅速,其技术创新与广泛应用推动了企业的数智化转型,带来了显著的商业价值。大模型得到了政府与社会各界的重视,党中央、国务院多次强调了人工智能技术在我国高质量发展中的重要作用。2023年以来,我国大模型产业迅速增长,大模型技术日渐成熟。然而,企业在应用大模型的过程中,除了遇到模型性能的问题,还会面临人员与组织管理等方面的难题。现有关于大模型的研究多聚焦于技术层面,缺乏商业管理视角的综合分析。本文通过调研四家实施大模型的企业,探讨了大模型商业化应用中的挑战与需求,并提出基于知识增强的大模型商业应用框架,为企业实践提供参考。
Since the release of ChatGPT,large language models(LLMs)have experienced explosive growth.LLMs'technological innovations and wide-ranging applications have unlocked tremendous potential for the digital and intelligent transformation of enterprises.In China,the development of AI and LLMs has garnered significant attention from the government,which have emphasized the strategic importance of domestic LLM in high-quality development.Since 2023,the domestic landscape of LLM development has witnessed a rapid proliferation of models,accompanied by significant advancements in core technologies.DeepSeek-known for its open-source accessibility,balanced capabilities,and low inference costs-has gained wide recognition both domestically and internationally.LLMs are becoming a driving force for localized innovation and application of AI.However,the path to commercializing LLMs remains fraught with challenges.Beyond ensuring model performance,enterprises face additional hurdles related to personnel training,organizational restructuring,and governance reform.Despite increasing scholarly attention,existing research largely centers on technical aspects,with limited systematic exploration from the viewpoint of enterprise management.This study adopts a qualitative methodology and conducts an in-depth analysis of four representative enterprises that have implemented LLMs.Through semi-structured interviews,document analysis,and cross-case comparisons,it systematically identifies the core challenges and obstacles faced in the process of commercial LLM deployment.The research focuses on how enterprises can overcome dual barriers in technology and management and adapt across key dimensions such as infrastructure,business processes,organizational structures,and human resources.Furthermore,the study offers a detailed examination of three mainstream application paradigms-prompt engineering,retrievalaugmented generation,and supervised fine-tuning with reinforcement learning from human feedback-clarifying their respective roles and implementation pathways in enterprise-level deployment.The findings indicate that successful commercialization of LLMs requires more than technological readinessit hinges on the effective integration of model capabilities with internal management strategies.Enterprises must simultaneously enhance cross-functional collaboration,strengthen data governance,and invest in workforce reskilling to fully unlock the transformative potential of LLMs.This study contributes a management-oriented framework that leverages domain-specific knowledge to guide the application of LLMs in business contexts.It identifies four essential pillars for implementation:infrastructure,user needs,model architecture,and deployment strategy.From a managerial perspective,the research outlines actionable strategies to help organizations navigate both organizational and technical challenges associated with embedding LLMs into core business operations.These insights offer a valuable reference for enterprises exploring AI-driven innovation and provide a new analytical lens for academic discussions on LLM commercialization.
作者
金旭磊
陈刚
黄丽华
肖帅勇
张成洪
JIN Xulei;CHEN Gang;HUANG Lihua;XIAO Shuaiyong;ZHANG Chenghong(School of Management,Fudan University,Shanghai 200433,China;School of Economics and Management,TongjiUniversity,Shanghai200092,China)
出处
《工程管理科技前沿》
北大核心
2025年第3期1-8,共8页
Frontiers of Science and Technology of Engineering Management
基金
国家自然科学基金专项资助项目(72342012)
国家自然科学基金资助项目(72271059)
国家自然科学基金重大资助项目(72394371)
国家自然科学基金青年资助项目(72301239,72301194)
国家社会科学基金重点资助项目(22AZD136)。
关键词
大模型
商业应用框架
案例分析
large language model
framework for commercial application
case analysis
作者简介
通讯作者:陈刚,复旦大学管理学院副研究员,博士,研究方向:大模型结构化推理。E-mail:chengang050970@foxmail.com。
