大语言模型的主权、安全及其治理

Sovereignty,Security and Governance of Large Language Models

ChatGPT的发布代表着人类在自然语言处理方面实现了质的飞跃,生成式人工智能成为主流发展方向。人工智能发展进入大语言模型军备竞赛阶段,这种竞赛不仅体现在企业之间,更存在于国家和地区之间,主权大语言模型的概念随之诞生并受到重视。作为一种新生事物,大语言模型会带来数据和网络、知识和信息、道德和伦理、公共和国家等方面的安全问题。这就需要树立多维和全息的安全观,并在观念和意识、策略和尺度、技术和方法等宏观、中观和微观层面综合治理和处置各种风险,以确保大语言模型的主权实现。

The release of ChatGPT represents a qualitative leap in human natural language processing,and artificial intelligence generated content has become the mainstream development direction.The development of artificial intelligence has entered the stage of a large language model arms race.This competition is not only reflected among enterprises,but also exists between countries and regions.The concept of sovereign large language models was born and received attention.As a phenomenon,large language models will bring security issues in data and networks,knowledge and information,morals and ethics,public and national security.This requires establishing a multi-dimensional and holographic security concept,manage and deal with various risks comprehensively at the macro-view,medium-view and micro-view such as concepts and awareness,strategies and scales,technologies and methods,so as to ensure the realization of large language model sovereignty.