摘要
大模型ChatGPT出现是否预示着“智能革命”?亟待对其进行系统性分析。本文首先从ChatGPT的创新效果评估出发,比较分析发现,在OpenAI的系列产品中属于渐进性创新和破坏式创新;而相较于同类竞品而言,确为突破性创新和颠覆性技术创新。而对比中国,ChatGPT事实上已经成为“卡脖子”技术体系的重要一环。继而进行技术体系解构式分析发现,作为“生产工具”,ChatGPT形成了人工系统领域的生产工具供给的范式革命;同时也实现了人工智能伦理治理的范式变革。追根溯源,ChatGPT产生的组织结构性要素恰恰来源于OpenAI的组织使命及围绕其构建的扁平化项目导向型组织和有限回报的融资机制创新,并在人工智能模型工程化过程中高效耦合开源与众包的研发模式。上述研究发现为后续关于ChatGPT引致的政治、经济、社会影响及其相关的技术治理夯实了研究基础。
Does ChatGPT signal an"intelligent revolution"?It is urgent to make a systematic analysis.In this paper,starting from the evaluation of the innovation effects of ChatGPT,the comparative analysis found that it belongs to progressive innovation and disruptive innovation in the series of products of OpenAI,Codex and InstructGPT for instance;ChatGPT is superior to Codex and InstructGPT in following human values,avoiding answers outside the scope of knowledge,dialogue ability,and in-task dialogue history modeling.The aforementioned conversational capability-related performance improvements come at the expense of some performance.n terms of code understanding and code generation,it is weaker than Codex;In terms of context learning ability,it is weaker than InstructGPT.Compared with similar competitive products,OPT of Meta,Sparrow of DeepMind,RL4LMs of Ai2,FLAN of Google,and so on,Chat-GPT is indeed a breakthrough innovation and subversive technological innovation.Within the Generative Pre-trained Model technolog-ical system of OpenAI,ChatGPT has formed a trans-boundary integrated disruptive technology evolution trajectory.In the entire AI e-cosystem including multimodal big models,it can be regarded as a performance transition type of disruptive technology.In contrast to a set of AI big models developed by the Chinese R&D startups,institutions or corporations,ChatCPT has become an important part of the:"bottleneck"technology system to China,including various sub-fields of artificial intelligence.With ChatGPT,USA has formed a certain technical"moat"in the subdivision field and the overall technical system,and there are inter-generational competitive advan-tages for Chinese large language models.It is found that ChatGPT,as a"production tool",forms a paradigm revolution in the supply of productivity tools in the field of artificial intelligence.As the disruptive technological innovation and one part of the basic technical cores of ChatGPT,Codex is manifested as a breakthrough in"production tools".The big model for coding generation,Codex essential-ly realizes two levels of innovation:functional product innovation and instrumental paradigm revolution.On the functional level,Codex and its subsequent series of large models gain the ability to understand code and complex reasoning,which in turn makes them more powerful than competing large models in understanding the practical effects of human language.On the productional level,Codex formed a paradigm revolution in the supply chain of code production tools in the field of artificial intelligence.The man-machine dia-logue mechanism of ChatGPT,in fact,provides a new technical solution for the ethical governance of artificial intelligence.ChatGPT has Continue the tradition of human-computer interaction,which always been an important way of artificial intelligence ethical govern-ance,in that chatbot has become an important technical solution of artificial intelligence ethical governance,and can unify the core ele-ments of technology ethical governance such as technology subject,technology object,technology itself,technology application process,and technology scene on the dialogue platform.So,at the same time,ChatGPT also realizes the paradigm change of AI ethical govern-ance.The organizational structural elements generated for ChatGPT can be traced back to the organizational mission of OpenAI,the flat project-oriented organization,the financing mechanism innovation with limited returns,and the efficient coupling of open source and crowdsourcing R&D organizational solution in the process of AI ModelOps.These findings lay a solid theorical and experiential founda-tion for further research on political,economic,and social impacts of the big models like ChatGPT and related technological govern-ance.
作者
张辉
刘鹏
姜钧译
曾雄
ZHANG Hui;LIU Peng;JIANG Jun-yi;ZENG Xiong(Shanghai Artificial Intelligence Laboratory,Shanghai 200232,China;Science and Technology Department of Hebei Province,Shijiazhuang 050021,China;National Natural Science Foundation,Beijing 100083,China;University of Science and Technology,School of Humans and Scocial Science,Beijing 100083,China)
出处
《科学学研究》
CSSCI
CSCD
北大核心
2023年第12期2113-2121,共9页
Studies in Science of Science
基金
科技创新2030-“新一代人工智能”重大项目(2022ZD0116205)
教育部人文社会科学研究青年基金项目(19YJCZH245)。
