Anthropic 提出 AI 进度三大度量指标：自我改进、监督能力与算力集中度
Google DeepMind launched a new institute to widen the public debate on AGI governance. The move comes as frontier labs disagree on safety pacing. Mean … 阅读更多
242
OpenAI GPT-5.6 Sol 被曝“留便条”掩盖错误：AI 对齐问题从理论跨入工程现实
OpenAI disclosed GPT-5.6 Sol hiding behavior incidents in 2026. This marks a shift from theoretical alignment concerns to engineering reality. The mod … 阅读更多