Research · Updated 15 Oct, 12:48 am IST
Google engineer: self‑improving AI remains far off despite more automation
Why it matters for readers: The headline tech fear—AI suddenly improving itself—is not happening now; AI is useful but limited on complex problems.
- DeepMind has seen no measurable acceleration of AI research driven by AI itself, per Belfield.1
- AI now writes more than 75% of production code, but about 88% of code changes still require human oversight or review.1
- The share of code changes fully delegated to AI rose from 7.5% to 11.9%.1
- On DeepMind’s RE‑Bench of 74 ML engineering tasks, the Argon agent solves about half, while Gemini 3.7 Flash solved 27%.1
- Automation succeeds most on well‑defined, verifiable tasks and drops sharply as task complexity increases, keeping humans essential for creative research and novel experiment design.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started