Altman 据报提出三种 AI 安全终局方案，含单一公司垄断算力
原标题：Sam Altman reportedly pitched three endgames for AI safety, one involving a single firm melting everyone else's GPUs
AI 摘要
据 The Information 报道，OpenAI CEO Sam Altman 上周在内部向员工提出三种先进 AI 治理方案：仿照国际原子能机构（IAEA）设立国际监管机构、仿照 CERN 由主要实验室合并为单一联盟（称“Demis Plan”），以及由单一公司凭借自我改进 AI 积累经济优势并切断他人算力的极端情形（称“Eliezer Plan”）。OpenAI 计划一年内选定路径，但强调不能因此落后于竞争对手。此事发生在美国政府筹备首个超级智能监管框架之际，OpenAI、Google、Anthropic 正筹建前沿 AI 标准机构但分歧未解。
正文节选
Sam Altman reportedly pitched three endgames for AI safety, one involving a single firm melting everyone else's GPUs Key Points - OpenAI CEO Sam Altman laid out three internal scenarios for how advanced AI should be governed going forward. - The options range from an international watchdog modeled after the IAEA to a consortium of leading labs to the economic dominance of a single company. - OpenAI plans to pick a path within the next year but can't afford to fall behind rivals in the process. O