Anthropic, OpenAI Warn on AI Self-Improvement Risks as Chinese Tech Giants Join Race
Anthropic and OpenAI have repeatedly warned since mid-September about the risks of失控 (loss of control) in Recursive Self-Improvement (RSI), urging a slowdown in R&D to allow time for safety measures. Chinese AI players are advancing aggressively: Zhipu's Tang Jie disclosed the company's first RSI engineering practice on September 17, Xiaomi released the MiMo-V2.6 series on September 22 as a "key step," and Alibaba's Wu Yongming said Qwen is exploring RSI. Anthropic、OpenAI警示AI自我改进风险 中美科技巨头竞逐RSI Anthropic和OpenAI自9月中旬以来多次警告递归自我改进(RSI)的失控风险，呼吁放缓研发进度以争取安全保障时间。中国AI企业加速布局：智谱唐杰于9月17日披露公司首个RSI工程化实践，小米于9月22日发布MiMo-V2.6系列称为“探索RSI的关键一步”，阿里吴泳铭透露Qwen团队正在RSI探索中。尽管存在警示，资本市场仍对RSI叙事反应积极。