招商银行首席信息官周天虹：银行使用人工智能技术一定要落实人的责任

Core Insights - The National Wealth Development Research Cooperation Platform Spring Summit 2025 focused on "Artificial Intelligence and the Future of Finance" [1] - The Chief Information Officer of China Merchants Bank highlighted the strengths and limitations of large language models, including issues like hallucinations, accuracy, alignment capabilities, value bias, ethical bias, and performance costs [1] Group 1: Model Management - The introduction of open-source models into banking requires a rigorous management process to determine which models can be used in production environments [1] - After introducing large language models, specific domain training is necessary, which must follow established protocols and be managed through appropriate systems and platforms [1] - Post-training evaluation of large language models is essential to ensure that any potential degradation in capabilities remains within acceptable limits [1] Group 2: Content Generation and Oversight - Supervision of content generated by large language models, particularly generative AI, is crucial, requiring both algorithmic and human oversight [2] - The financial industry emphasizes the importance of accountability, mandating that responsibility is assigned to individuals in key areas and processes when using AI technologies [2]