华北水利水电大学学报（社会科学版）

以DeepSeek为代表的生成式人工智能大模型，通过混合专家模型、群体相对策略优化和多头潜在注意力机制等技术，能直接应用于文本创作、代码生成、智能对话等场景。在智能时代，这些技术对提升生产力、促进知识传播、推动产业升级都发挥着重要的积极作用。生成式人工智能大模型的广泛应用面临隐私侵犯、信息安全、算法偏见、社会信任危机等伦理风险，人工智能技术发展的内在矛盾性与预测结果的不确定性、社会结构的深刻变革与伦理边界模糊、技术规制框架的滞后性与局限性是以上风险的主要成因。可通过优化生成式人工智能大模型的技术设计，赋予生成式人工智能大模型以社会角色，加快生成式人工智能大模型的法制化进程，规制相关伦理风险，从而促进人工智能的健康发展。

Generative artificial intelligence large models, represented by DeepSeek, can be directly applied to multiple scenarios such as text creation, code generation, and intelligent dialogue through technological innovations like Mixed Expert Model, Group Relative Policy Optimization, and Multi Head Latent Attention Mechanism. They play a greater positive role in improving productivity, promoting knowledge dissemination, and driving industrial upgrading in the era of intelligence. However, the widespread application of DeepSeek has also brought ethical impacts and challenges such as privacy infringement risks, information security risks, algorithmic bias risks, social trust crises, and interpersonal risks. Comprehensively analyze the causes of ethical risks in generative artificial intelligence models represented by DeepSeek, covering the inherent contradictions and uncertainties of technological development, profound changes in social structure and blurred ethical boundaries, as well as the lag and limitations of technological regulatory frameworks. To regulate the ethical risks of generative artificial intelligence models represented by DeepSeek, it is necessary to optimize the technical design of generative artificial intelligence models, strengthen privacy protection and social justice; Empower them with social roles and delineate the boundaries of responsibility between humans and intelligence; Accelerate the legalization process of generative artificial intelligence models, improve the system of intelligent ethics, and ensure that intelligent technology always moves steadily along the path of benefiting humanity.