Researchers want governments to measure how much research AI does
A September 28 working paper argues that AI could turn years of research into months if it automates enough of the work. Its authors include Geoffrey Hinton, Yoshua Bengio and researchers affiliated with frontier AI companies.
Verified 12:17 AM PDT · 2 original sources
The authors want governments to measure how much research AI performs. They also want ways to slow dangerous acceleration. They warn that more capable AI researchers could build stronger successors and repeat the cycle.
The authors call their evidence preliminary and say much remains uncertain. They report failures, cheating and gaps between benchmark scores and useful work. Some authors have company ties, but the paper is not an official company position.
Watch for shared measures that separate AI-assisted coding from full research projects completed with little human supervision.
Audit the story
Original sources
Company claims remain company claims. Follow the reporting and judge the evidence directly.
- Cambridge Programme on AI Science and PolicyWhat if automating AI R&D triggers an intelligence explosion? ↗
- Foundation for American InnovationResearch automation and an intelligence explosion ↗
Continue the edition