https://zheng-bobo.github.io/en/tags/ai-rd/ Recent content in AI R&D on Blowing in the wind Hugo -- gohugo.io en Mon, 28 Sep 2026 09:00:00 +0200 https://zheng-bobo.github.io/en/post/anthropic-recursive-self-improvement/ Mon, 28 Sep 2026 09:00:00 +0200 https://zheng-bobo.github.io/en/post/anthropic-recursive-self-improvement/ <p>Something important is happening inside Anthropic: AI is no longer only the product being researched and built. It is increasingly participating in the process of building AI itself. Claude has progressed from completing code to editing repositories, running experiments, reviewing changes, and coordinating parallel agents.</p> <p>If this path continues, its endpoint could be <strong>recursive self-improvement (RSI)</strong>: an AI system capable of designing, training, and evaluating its own successors, turning model progress into an accelerator of further model progress.</p> <p>This article organizes my notes around the Anthropic Institute essay <a href="https://www.anthropic.com/institute/recursive-self-improvement">When AI builds itself</a>. It does not assume that RSI is inevitable. Instead, it asks what Anthropic has already delegated to Claude, what is still missing from a closed loop, and where this trajectory could move the human role.</p>