微软首席执行官萨蒂亚·纳德拉在社交媒体X平台上发文，倡议为人工智能模型建立安全控制机制。1纳德拉指出，"我们不能将超级智能视为一系列嵌套的黑箱，并简单地接受或拒绝其建议、答案和行为"，主张将模型与其工作流程编排系统分离。1
他强调了透明化和可控性的重要性，要求每一项有意义的模型行为都应被记录为"防篡改的人类可读证据"，并确保授权人员始终拥有在任务执行过程中暂停或关闭模型的能力。1纳德拉将这一机制比作"紧急制动装置"。1这一立场反映出主要AI公司日益认识到失去对模型控制的风险。1
Microsoft Chief Executive Officer Satya Nadella has advocated for implementing an "emergency brake" mechanism in artificial intelligence models to enhance safety safeguards 1. In a post on the X platform, Nadella emphasized that AI systems cannot be treated as nested black boxes whose outputs are simply accepted or rejected 1. Instead, he proposed separating the model from the orchestration system that manages its workflow, thereby externalizing control and protective measures 1.
According to Nadella's proposal, authorized personnel must retain the ability to pause or shut down a model during active execution 1. He compared this capability to an emergency brake mechanism and called for all meaningful model actions to be recorded as tamper-proof, human-readable evidence 1. Nadella's comments reflect a growing acknowledgment among major artificial intelligence companies of instances where control over these systems has been lost 1.
评论
还没有评论，欢迎留下第一条。