Sakana 的 RSI 实验室押注 AI 自我改进以打破算力军备竞赛
sakana ai 推出了 rsi 实验室，致力于开发能够编写自己代码的递归式自我改进智能体，这提供了一条挑战前沿实验室重度依赖 gpu 缩放的高效 ai 路径。
sakana ai introduced a research division dedicated to recursive self-improvement, outlining a four-phase roadmap toward agents that autonomously optimize their technical foundations and code. building on prior systems like the ai scientist and darwin gödel machine, the lab aims to demonstrate that algorithmic efficiency can rival the performance gains from scaling compute. this strategy offers a potential alternative to the dominant scaling paradigm, suggesting that iterative self-improvement could lower barriers to frontier ai development.