Comprehensive Impact Analysis of GPT-4: High-quality Economic Development and National Security Prevention
-
摘要: 人工智能自诞生以来历经多次迭代演进，现已成为世界经济和社会变革的重大驱动力量。大型语言模型ChatGPT的问世及其迭代版本多模态GPT-4的发布，颠覆了人类对生成式AI的认知，引发全球广泛关注。在全面建设社会主义现代化国家的重要起点和关键阶段，我国的主要目标是经济高质量发展取得新突破。GPT-4代表的AI算法和模型技术在生产、金融、营销、教育、医疗、外贸和投资等领域均具有广阔应用前景，未来必将成为我国经济高质量发展的重要抓手，赋能更多行业和场景。但以GPT-4技术为代表的人工智能应用也会给国家经济、政治、法律和社会等领域带来诸多安全风险与挑战，须采取系列安全防范策略掌控好AI应用风险，为我国经济高质量发展与国家安全保驾护航。Abstract: Since the advent of artificial intelligence (AI), it has now become a major driving force for world economic and social change after having undergone many iterations and evolution. The advent of large-scale language model ChatGPT and its multimodal GPT-4 iteration has subverted human cognition of generative AI and attracted worldwide attention. At the important starting point and key stage of building a socialist modern country in an all-round way, China's main goal is to achieve new breakthroughs in high-quality economic development. The AI algorithm and modeling techniques represented by GPT-4 have a wide application prospect in fields such as production, finance, marketing, education, healthcare, foreign trade and investment, and will become an important lever for China's high-quality economic development in the future, empowering more industries and application scenarios. At the same time, AI applications represented by GPT-4 technology will also bring security risks and challenges to many fields such as national economy, politics, law and society. Therefore, a series of security prevention strategies must be adopted to control AI application risks and escort China's high-quality economic development and national security.
-
Key words:
- artificial intelligence /
- ChatGPT /
- GPT-4 /
- high-quality development /
- national security
-
表 1 GPT系列模型一览表
模型名称 发布时间 参数数量(亿) 数据集 数据量 训练方法 目前局限性 GPT 2018.6 1.17 BooksCorpus和英文维基百科 5GB 基于Transformer模型，采用自回归方式预训练 只能处理单语言文本，不能较好地适应新任务 GPT-2 2019.2 15 WebText(即社交媒体平台Reddit上高赞文章) 40GB 基于深层Transformer模型，采用自回归方式预训练，能够输出流畅文本 只能处理单语言文本，可能输出存在偏见或不真实的内容 GPT-3 2020.5 1750 Common Crawl等多个数据源(包括WebText) 45TB 基于深层Transformer模型，采用自回归方式预训练，能完成多种NLP任务 只能处理单语言文本，可能输出存在偏见或不真实的内容，需要大量计算资源和存储空间 ChatGPT(GPT-3.5) 2022.11 未公布 未公布(包括文本代码混合体) 未公布(可能超过50TB) 基于深层Transformer模型，采用自回归方式预训练，能接受文本输入，可进行流畅对话 可能输出存在偏见或不真实的内容，需要大量计算资源和存储空间 GPT-4 2023.3 未公布 未公布 未公布(可能超过100TB) 基于深层Transformer模型，采用自回归方式预训练，能接受文本和图像输入，可进行流畅对话 可能输出存在偏见或不真实的内容，需要大量计算资源和存储空间 注: 此表由作者根据相关资料整理得出①。 
下载: 导出CSV
-
[1] 晓镜. SK电讯剑指"超级应用"[N]. 人民邮电报, 2023-03-12(07). [2] 张佳欣, 刘园园, 陈曦, 等. "顶流"之下, 看人工智能喜与忧[N]. 科技日报, 2023-02-16(005). [3] BUCHANAN B G. The history of artificial intelligence[J]. AI Magazine, 2005, 26(4): 9-21. [4] DREYFUS H L. What computers can't do: a critique of artificial reason[M]. New York: Harper & Row, 1972: 630-631. [5] 张夏恒. 新一代人工智能技术(ChatGPT)及其对人类社会的影响与变革[J/OL]. 产业经济评论, 2023-03-10(001), https://kns.cnki.net .[6] 沈威. ChatGPT: 形成机理与问题应对[N]. 中国社会科学报, 2023-03-07(007). [7] KAREN W. ChatGPT is poised to change medical care[N]. USA Today, 2023-03-06. [8] CASCELLA M, MONTOMOLI J, BELLINI V, et al. Evaluating the feasibility of ChatGPT in healthcare: an analysis of multiple clinical and research scenarios[J]. Journal of medical systems, 2023, 47(1): 33. doi: 10.1007/s10916-023-01925-4 [9] HOMOLAK J. Opportunities and risks of ChatGPT in medicine, science, and academic publishing: a modern promethean dilemma[J]. Croatian Medical Journal, 2023, 64(1): 1-3. doi: 10.3325/cmj.2023.64.1 [10] LUND B D, WANG T, MANNURU N R, et al. ChatGPT and a new academic reality: artificial intelligence-written research papers and the ethics of the large language models in scholarly publishing[J]. Journal of the Association for Information Science & Technology, 2023, 1: 1-12. [11] ROOSE K. Once the future, has become the present. What do we do now?[N]. New York Times, 2023-01-13(B1-B5). [12] HUANG K. AI is doing homework. Can it be outsmarted?(cover story)[N]. New York Times, 2023-01-17(A1-A16). [13] STOKEL-WALKER C. AI bot ChatGPT writes smart essays - should professors worry?[J]. Nature, 2022-12-9(Date of Electronic Publication). [14] 郭小平, 秦艺轩. 解构智能传播的数据神话: 算法偏见的成因与风险治理路径[J]. 现代传播(中国传媒大学学报), 2019(9): 19-24. https://www.cnki.com.cn/Article/CJFDTOTAL-XDCB201909005.htm [15] SIEGERINK B, PET L. ChatGPT as an author of academic papers is wrong and highlights the concepts of accountability and contributorship[J]. Nurse education in practice, 2023, 68: 103599. doi: 10.1016/j.nepr.2023.103599 [16] STOKEL-WALKER C, VAN NOORDEN R. What ChatGPT and generative AI mean for science[J]. Nature, 2023, 614: 214-216. doi: 10.1038/d41586-023-00340-6 -
