Anthropic 首次公开呼吁暂停前沿 AI 开发，并披露内部 RSI 数据，做 AI 安全或政策研究的团队值得关注——这可能是行业转折点的信号。
Anthropic 在最新博文中提出，世界应有权选择减缓或暂停前沿 AI 的发展。文章聚焦于递归自我改进（RSI），即 AI 自主设计并开发下一代 AI 的能力。Anthropic 透露，公司内部已出现 RSI 迹象：超过 80% 的合并代码由 Claude 编写，而 2025 年初 Claude Code 推出前这一比例仅为个位数。典型工程师每季度交付的代码量是过去的 8 倍。关于暂停，Anthropic 指出关键在于协调和验证全球竞争对手是否真的停止开发，而建立类似军控条约的基础设施和信任需要数十年，但时间并不充裕。
"We believe it would be good for the world to have the option to slow or temporarily pause frontier ...
"We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development." A jarring quote from Anthropic in today's 'When AI builds itself' post on recursive self-improvement (RSI). The piece lays out where the tech is heading: AI that autonomously designs and develops its own successor. Anthropic shared the signs of RSI already being seen within the company. More than 80% of the code the lab merges is now written by Claude, up from low single digits prior to Claude Code launching in early 2025. The typical engineer also ships 8x as much code per quarter as they used to. On a potential pause, Anthropic said the concern is coordination and actually proving rivals across the globe have stopped development. The arms-control treaties that did this "took decades to build both the infrastructure and the trust." "We don't have that long." 💬 8 🔄 1 ❤️ 6 👀 1335 📊 8 ⚡