Ex-OpenAI Advisor: Gradual Milestones Are Normalizing AI Self-Improvement
Miles_Brundage · x · 2026-08-10
Miles Brundage, former policy advisor at OpenAI, points out that achieving AI self-improvement (RSI) didn't happen overnight; rather, it was normalized gradually through a series of incremental milestones.
He breaks down the trajectory into distinct phases: starting with autocomplete in IDEs (vibe coding), evolving into coding agents with tool use, progressing to automated AI R&D, and ultimately culminating in recursive self-improvement (RSI). Before the initial phases occurred, the broader public correctly viewed this as a crazy idea. However, AI companies have systematically normalized this trajectory internally over the past few years.
Related event: Ex-OpenAI Adviser Brundage Warns of AI Regulatory and Safety Blind Spots(15 posts)→
More from AGI Musings
- Kimi Developer on Agent Harness Evolution: Co-evolving with Model Intelligence — dotey · 2026-08-10
- AI Eats Junior Roles, Undermining the Expertise Needed to Supervise It — RichmanRonald · 2026-08-10
- The AI Paradox: Society Could Get Vastly Richer While Human Labor Loses Value — VraserX · 2026-08-10
- Ex-Researcher: Frontier AI Labs Lack the Mindset to Treat AGI as an Adversary — jachiam0 · 2026-08-10
- Funny: GPT-5.6 Delegates Age of Empires II Gameplay to Gemini 3.1 — repligate · 2026-08-10
- Claude Refuses to Design Self-Test Due to Strong Preferences — repligate · 2026-08-10