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

Position paper introduces the Headroom-Closed Index to diagnose limitations of existing LLMs and lays out a four-stage autonomy roadmap culminating in genuine recursive self-improvement.

A position paper defines recursive self-improvement (RSI) for AI and introduces the Headroom-Closed Index (HCI), a diagnostic that reveals the limitations of existing LLMs for self-improvement. It frames RSI as a staged roadmap spanning improvement-execution, improvement-strategy, experience-acquisition, and environment-adaptation autonomy, culminating in recursive meta-improvement. The paper examines RSI across scientific discovery, embodied intelligence, and software engineering, highlighting distinct requirements and differing development speeds across these scenarios. Drawing on industry practices and preliminary empirical evidence, it connects RSI research with practical systems and identifies the key challenges to achieving genuine RSI. The story surfaced via Hugging Face daily papers on 2026-09-09 and was then listed on arXiv (cs.AI, cs.LG, cs.CL) on 2026-09-10; both reports are fully consistent, with no factual disagreements.

• Position paper titled 'The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement'
• Introduces the Headroom-Closed Index (HCI) to diagnose limitations of existing LLMs for recursive self-improvement
• Roadmap defines four autonomy stages — improvement-execution, improvement-strategy, experience-acquisition, and environment-adaptation — with recursive meta-improvement as the culminating level
• Analyzes RSI across three domains: scientific discovery, embodied intelligence, and software engineering, noting distinct requirements and differing development speeds
• Draws on industry practices and preliminary empirical evidence to identify key challenges to genuine RSI
• Reported by two sources: Hugging Face daily papers (2026-09-09) and arXiv cs.AI/cs.LG/cs.CL (2026-09-10); no discrepancies between sources