AI Researcher: Recursive Self-Improvement Bottlenecked by Ecosystem Data
herbiebradley · x · 2026-07-24
AI researcher Herbie Bradley offers a fresh perspective on recursive self-improvement (RSI) in a recent interview. He argues that RSI should be viewed not as a loop closing within a single lab, but as a loop that closes across the entire economic ecosystem.
Bradley emphasizes that the biggest obstacle to AI self-improvement is the data bottleneck. Even if an AI reaches the proficiency of a top-tier researcher like Ilya Sutskever, training it to become an expert in entirely different domains, such as investment banking or law, is severely constrained by the lack of available training data. He notes that RL alignment companies and data vendors are expending massive efforts to source missing data, yet many enterprises remain reluctant to sell their proprietary data.
More from AGI Musings
- Frontier AI labs are shifting their messaging on jobs and AI impact — menhguin · 2026-07-24
- Scott Galloway says cheaper Chinese open-weight models could break US AI valuations — r0ck3t23 · 2026-07-24
- AI researcher says one hard task can block full automation of entire jobs — herbiebradley · 2026-07-24
- From Builders to Debuggers: The Shift in AI-Era Software Development — zakelfassi · 2026-07-24
- AI labs sell recursive self-improvement because 8% progress sounds too small — joshalbrecht · 2026-07-24
- a16z's Martin Casado: Easier to Teach Systems People AI Than AI People Systems — mgill25 · 2026-07-24