《计算机应用》唯一官方网站

• •    下一篇

大语言模型的技术应用前景与风险挑战

徐月梅1,胡玲1,赵佳艺1,杜宛泽1,王文清2   

  1. 1. 北京外国语大学
    2. 北京大学
  • 收稿日期:2023-07-06 修回日期:2023-07-29 发布日期:2023-09-14 出版日期:2023-09-14
  • 通讯作者: 徐月梅
  • 基金资助:
    面向多语种社交新闻的跨语言情感分析研究

Technology application prospects and risk challenges of large language model

  • Received:2023-07-06 Revised:2023-07-29 Online:2023-09-14 Published:2023-09-14

摘要: 摘 要: 针对大语言模型(LLM)技术的快速发展,剖析其技术应用前景及风险挑战,对通用人工智能(AGI)的发展和治理有重要参考价值。首先,以BERT(Bidirectional Encoder Representations from Transformers)、GPT(Generative Pre-trained Transformer)和ChatGPT(Chat Generative Pre-Trained Transformer)等语言模型为代表,综述大语言模型的发展脉络、大语言模型的核心技术和评估体系;然后,分析大语言模型现存的技术局限和安全风险;最后,提出大语言模型在技术上改进、政策上跟进的建议。分析指出作为发展阶段的大语言模型,现有模型存在非真实性及偏见性输出、实时自主学习能力上尚有欠缺,算力需求庞大,对数据质量和数量依赖性强,语言风格单一;存在数据隐私、信息安全、伦理等方面的安全风险。未来发展可从技术上继续改进,从“大规模”转向“轻量化”、从“单模态”走向“多模态”、从“通用”迈入“垂类”;从政策上实时跟进,实施有针对性的监管措施,规范其应用和发展。

关键词: 大语言模型, 风险挑战, 技术监管, 应用前景, 通用人工智能

Abstract: Abstract: In view of the rapid development of large language model (LLM) technology, a comprehensive analysis was conducted on its technical application prospects, risks and challenges which has great reference value for the development and governance of artificial general intelligence (AGI). Firstly, with representative language models such as Multi-BERT (Multilingual Bidirectional Encoder Representations from Transformer), GPT (Generative Pre-trained Transformer) and ChatGPT (Chat Generative Pre-Trained Transformer), the development, core technologies and evaluation system of LLM were reviewed. Then, a detailed analysis of LLM on technical limitations and security risks was conducted. Finally, suggestions were put forward for technical improvement and policy follow-up of the LLM. The analysis indicates that as a developing status, the current LLM still produce non-truthful and biased output, lack real-time autonomous learning ability, require huge computing power, highly rely on data quality and quantity, and tend towards monotonous language style. There are security risks related to data privacy, information security, ethics, and other aspects. Future developments can continue to improve technically, from "large-scale" to "lightweight", from "single-modal" to "multi-modal", from "general-purpose" to "vertical"; For real-time follow-up in policy, targeted regulatory measures should be implemented to regulate their applications and developments.

Key words: Keywords: Large Language Model (LLM), risk challenge, technology supervision, application prospect, Artificial General Intelligence (AGI)