22 Researchers Examine Recursive Self-Improvement in AI Research
A paper titled “What if automating AI R&D triggers an intelligence explosion?” was coauthored by 22 researchers, including Geoffrey Hinton, Yoshua Bengio, Andrew Barto and OpenAI chief scientist Jakub Pachocki. The paper models a scenario in which AI fully participates in AI research, creating a recursive feedback loop. Under assumptions including full automation, sustained research returns and no new compute or data bottlenecks, the model estimates that AI development could become up to 10 times faster in roughly 18 months. The paper and related reporting emphasize that this is a conditional scenario, not a measured result or consensus forecast. Current evidence does not show that an intelligence explosion has occurred or that the feedback loop is self-sustaining; reliability, compute, data, training time, research taste and diminishing returns remain constraints.
< img id="wx_img" src="https://www.qbitai.com/wp-content/uploads/imgs/qbitai-logo-1.png" width="400" height="400">
2026-10-05 12:42:52 来源:量子位
AI已经开始真正进入「造下一代AI」的流水线
梦瑶 发自 凹非寺
量子位 | 公众号 QbitAI
当AI开始参与研发下一代AI,人类可能已经站到「智能爆炸」前夜。
#Geoffrey Hinton#Yoshua Bengio#Andrew Barto#Jakub Pachocki#Sören Mindermann#Alan Chan#OpenAI#Anthropic#Claude#Microsoft#UC Berkeley#Tsinghua University