Google says self-improving AI is nowhere close

Google engineer Haydn Belfield says DeepMind data shows no measurable acceleration of AI research by AI, and self-improving AI is far off. AI writes over 75% of production code, yet 88% of changes still need human review; the in-house General Capability Index shows steady, not sudden, growth.

• Kevin Kelly's 2016 AI predictions, 10 years later2026-10-12
• Google launches Gemini 4 Argon frontier model2026-09-30
• Ex-DeepMind researcher raising funds for AI reasoning startup2026-09-24
• Google DeepMind's Dream-RSI lets AI agents improve by 'dreaming'2026-09-19
• Ex‑DeepMind PR says lab banned public talk of extinction risk2026-09-10