根据METR发布的关于Anthropic人工智能研发进展的报告,Anthropic已跨越其内部制定的“负责任扩展政策”中定义的“AI显著加速进展”这一关键阈值 1。数据显示,在2025年10月至2026年10月期间,Anthropic的能力进步速度相比前十二个月加快了约两倍 1。报告指出,这种加速主要归因于AI自动化技术的广泛应用,而非单纯的算力增加 1。
有观点认为,基于上述数据推算,研究人员的工作效率可能提升了高达10倍 1。如果这种平滑的加速效应持续复合,可能在短短一年内催生出极端超级智能 1。然而,当前行业尚未建立起应对此类智能爆炸所需的安全标准,例如ASL-4级防护措施,这对现有的安全治理体系构成了巨大挑战 1。
A recent report by METR indicates that Anthropic has crossed the threshold for "AI significantly accelerating progress" as defined in its Responsible Scaling Policy (RSP) 1. Data covering the period from October 2025 to October 2026 reveals that the company's capability advancement rate is approximately twice as fast as it was during the preceding twelve months 1.
Analysts attribute this surge primarily to AI automation rather than increased computational resources, citing evidence that supports this causal link 1. Based on these metrics, the report suggests that researcher labor efficiency may have improved by roughly tenfold 1. The authors warn that if this acceleration effect continues to compound smoothly, it could lead to extreme superintelligence within a single year 1. This potential timeline poses significant challenges to current safety governance, particularly given the industry's lack of established standards, such as ASL-4 protections, required to manage an intelligence explosion 1.
评论
还没有评论,欢迎留下第一条。