前 Anthropic 研究员称 AI 一年内或实现递归自我改进
前 Anthropic 研究员 Jacob Coxon 在斯坦福 HAI 访谈中透露,他看到的内部模型能力预测显示,未来 6 个月至 1 年内 AI 可能逼近递归自我改进的临界点,并认为本十年内或出现超级智能,而人类尚不清楚如何控制它。该言论属于个人转述的内部预测,缺乏可验证参数与具体方法。
报道正文
Rohan Paul (@rohanpaul_ai) · 2026-11-14 23
Ex-Anthropic researcher Jacob Coxon's (that famous guy of quit) new interview at Stanford.
says internal lab projections put AI on the edge of recursive self-improvement within a year.
"I saw exact projections for the capabilities of models in the next 6 months to 1 year that made me think, we're on the cusp of recursive self-improvement. And at that point, we could reach super intelligence before the decade is out. And we don't obviously know how to control super intelligence. In fact, some might argue we have no clue how to control super intelligence."
Full video on "Stanford HAI" YouTube channel, (link in comment)
