大模型技术发展趋势及应用  

Development Trend and Application of Large Model Technology

作　　者：屠要峰 黄卫东[2,3] Tu Yaofeng;Huang Weidong(ZTE Center Research Institute,Nanjing 210012,China;Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Information Industry Development Strategy Institute,Nanjing 210003,China)

机构地区：[1]中兴通讯中心研究院,南京210012 [2]南京邮电大学,南京210003 [3]信息产业发展战略研究院,南京210003

出　　处：《信息通信技术》2024年第3期42-49,共8页Information and communications Technologies

摘　　要：大模型的出现和发展为人工智能领域带来了前所未有的活力与机遇。文章首先对人工智能大模型的发展历程进行了剖析,从人工智能的符号主义到连接主义的演进,再到深度学习的发展和大模型的崛起。然后,综合分析了大模型在技术方面存在的挑战及业界在长上下文技术、混合专家(MoE)技术、多模态技术以及非Transformer架构等方面的创新,进一步探讨大模型智能体及其在行业领域的应用前景。最后对大模型的未来进行了展望并提出发展策略。The emergence and development of large language models(LLMs)have brought unprecedented vitality and opportunities to the field of artificial intelligence.Firstly,the article analyzes the development process of LLMs and artificial intelligence,tracing the evolution from symbolic AI to connectionism,and then to the rise of deep learning and LLMs.It then comprehensively examines the technical challenges of LLMs and the industry's innovations in long-context techniques,mixture of experts(MoE)techniques,multimodal technologies and non-Transformer architectures.Furthermore,the article explores the prospects of LLM agents and their applications in various industry sectors.Finally,it looks ahead to the future of LLMs and proposes development strategies.

关 键 词：大语言模型 自注意力机制 混合专家 多模态 大模型智能体 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.1[自动化与计算机技术—控制科学与工程]