Core Insights - The collaboration between Dongbi Technology Data and Shanghai University of Finance and Economics has led to the establishment of China's first high-quality AI governance technology corpus and a multi-governance decision support model, marking a significant step in exploring efficient and collaborative AI governance models [1][2] Group 1: AI Governance Challenges - The rapid evolution of AI technology has highlighted issues such as data security, algorithmic bias, ethical lapses, employment impacts, and potential governance gaps, making collaborative governance a crucial industry consensus for orderly AI development [1] - The "Global AI Governance Action Plan" emphasizes the need for timely risk assessment and the establishment of a widely accepted safety governance framework [1] Group 2: Development of AI Governance Corpus - Dongbi Technology Data has created a high-quality AI governance technology corpus, focusing on 14 types of governance risks, including backdoor attacks and data poisoning, by collecting over 500 high-quality English journal papers and more than 1,500 core Chinese journal papers [2] - The corpus also integrates over 1,000 high-quality normative texts, including laws, regulations, policy documents, and case studies from 18 ministries and 16 local government departments [2] Group 3: AI Multi-Governance Decision Support Model - The newly developed AI multi-governance decision support model focuses on five core tasks: knowledge Q&A, case inquiry and analysis, technical solution consultation, governance plan generation, and resource search [3] - The model has been fine-tuned using over 2,000 high-quality Q&A pairs, achieving an accuracy rate of 91.4% and a hallucination rate of only 1.5% in a test set of 1,000 governance-related queries [3] - Future plans include continuous updates to the AI governance corpus and gradual opening of the decision support model to enterprises and government departments to enhance AI governance capabilities [3]

"这一覆盖学术前沿与政策实践的多维科技语料，将为推进我国人工智能协同治理工具的研发提供关键 数据支撑。"吴登生说。 在此基础上，上海财经大学数字经济学院与东壁科技数据联合开发了全国首个人工智能多元共治决策支 持大模型，为人工智能协同治理提供关键工具支撑。 据悉，这一模型专注于人工智能治理领域知识问答、案例查询与分析、技术方案咨询、治理方案生成、 资源查找五大核心任务，在国产开源大语言模型基础上，利用构建的2000多条高质量人工智能治理领域 问答对语料进行精细化微调，使其能够严格遵循人工智能治理领域回答范式。 此外，这个模型还接入了基于前述人工智能治理科技语料库构建的可持续动态更新数据库，通过检索增 强生成（RAG）技术，模型能够有效引用权威语料库内容，从而大幅提升专业回答的准确性和可靠 性，显著缓解通用模型在这一领域易产生"幻觉"的风险。 新华财经上海7月29日电（记者潘清）28日举行的2025世界人工智能大会"大模型治理和有序发展生态论 坛"上，东壁科技数据联合上海财经大学数字经济学院共同构建的高质量人工智能治理科技语料和人工 智能多元共治决策支持大模型正式发布。 校企携手发布两项成果，将为人工智能协同治理 ...