RSI, AI's New Buzzword: Recursive Self-Improvement or a Fear-Driven Business?
APPSO · wechat · 2026-09-15
After AGI, the AI community is hyping a new buzzword: RSI (Recursive Self-Improvement), where AI writes its own code, runs experiments, and improves training pipelines for each successive model generation. The piece traces the concept back to I.J. Good's 1965 "intelligence explosion" idea and a recent preprint from SJTU, Tsinghua, ByteDance and others proposing an L1–L5 autonomy scale.
On the practical side: OpenAI says it has hit its "automated research intern" goal, with agent runtime equal to 3.1 human workdays per day invested; Anthropic used Claude Sonnet 5 to improve Claude Opus 4.8's alignment in 60 hours with only 2,400 training samples; and Lilian Weng returned to OpenAI to lead recursive self-improvement research. Yet ASI timeline predictions range widely (3–10 years), and catastrophic forgetting plus research judgment remain unsolved hurdles.
The article's core argument targets the doomsday business logic: Jacob Coxon's resignation from Anthropic sparked the safety debate, Amodei called for a slowdown with third-party auditors, and even Altman and Musk aligned on "slowing down" — but only big labs can afford such audits, and they write the rules. Huang, defending GPU sales, rejects the safety-vs-innovation framing. Ultimately, the author argues both doom and anti-doom camps are fighting over how compute, capital and industry profits get divided.
Related event: AI circles hype RSI: intelligence explosion or fear-mongering business(2 posts)→
More from AGI Musings
- Nat Lambert: Open models could be what holds the AI ecosystem together — natolambert · 2026-10-31
- Kai-Fu Lee's 'AI Native' ships Nov 3 with endorsement from Satya Nadella — kaifulee · 2026-10-31
- AI security's dark age: agents can probe every endpoint and attack surface — graceisford · 2026-10-31
- AI dinner takeaways: orgs shipping 5x faster, yet GDP isn't growing 30% a year — graceisford · 2026-10-31
- Companies that replaced engineers with AI are rehiring seniors at a premium — DavidLinthicum · 2026-10-31
- Nathan Young proposes prediction markets and AI reporting for AI-community accountability — NathanpmYoung · 2026-10-31