The Lab That Started Eating Its Own Tail
METR, an independent nonprofit, spent 11 weeks inside Anthropic and found its models improved about twice as fast over the past year - and roughly 80 percent of that acceleration came from using AI to build AI. Why it
METR, an independent nonprofit, spent 11 weeks inside Anthropic and found its models improved about twice as fast over the past year - and roughly 80 percent of that acceleration came from using AI to build AI.
Why it matters
- AECI acceleration since Oct 2025: 1.9x (95% CI 1.5-2.8)
- Trend slope went from ~15 to ~29 points per year
- The speed-up was NOT from more training compute or headcount
- Caveat: Anthropic redacted most of the evidence, so the public must trust METR's judgment
The machine is now helping build the next machine - and there is no visible bottleneck for 12 months.
Full story: https://www.threerouter.com/blog/333
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.