The Last AI Built by Humans: Towards Genuine Recursive Self-Improvement

Work

by Yi Duan, Ying Liu, Zirui Tang, Haodong Chen, Jun Zhou, Yumou Liu · 2026

This paper examines recursive self-improvement (RSI), in which AI systems use experience and feedback to make persistent changes that improve both their capabilities and the process of future improvement. It introduces the Headroom-Closed Index to characterize limitations of current large language models and proposes a five-stage autonomy roadmap spanning improvement execution, strategy, experience acquisition, environment adaptation, and recursive meta-improvement, with discussion of applications and challenges across scientific discovery, embodied intelligence, and software engineering.