Google 称自我改进型 AI 仍遥不可及
原标题：Google says self-improving AI isn't anywhere close
AI 摘要
Google DeepMind 工程师 Haydn Belfield 表示，目前没有可测量的证据表明 AI 正在加速 AI 自身研究，自我改进型 AI 仍遥不可及。尽管 AI 已编写超过 75% 的生产代码，但约 88% 的代码变更仍需人工审核，完全委托给 AI 的变更比例仅从 7.5% 升至 11.9%。在 1,389 次内部 Argon 智能体运行中，任务越复杂，无人工干预的成功率下降越明显；DeepMind 自研基准 RE-Bench 上 Argon 仅解决约一半任务。Google 的通用能力指数显示 Gemini 能力以每月约 0.1 分稳定增长，未见加速。该结论与 Geoffrey Hinton、Yoshua Bengio 等 20 多位研究者近期警告的“智能爆炸”观点相左。
正文节选
Google says self-improving AI isn't anywhere close Google engineer Haydn Belfield says self-improving AI remains far off, even as significant automation is already taking hold. Google Deepmind has seen no measurable acceleration of AI research driven by AI itself, according to Belfield. That doesn't mean significant automation isn't happening already: AI now writes more than 75 percent of production code. But about 88 percent of code changes still need human oversight or review, and every change