Demis Hassabis on the New Coming Age

Zvi (Don't Worry About the Vase) · 2 days ago ● 3 sources

Google CEO Demis Hassabis proposed a Frontier AI Standards Body within the U.S. Government to evaluate and oversee frontier AI models, with voluntary pre-release reviews by companies 30 days before deployment. Critics including AI safety researchers argue the proposal is insufficient, noting that disagreement among experts ranges from 5% to 90% regarding catastrophic risks, and that internal model development would escape oversight. The post also covers DeepMind's bioresilience framework and Alex Turner's resignation over Google's decision to provide AI models to the Department of War without restrictions on autonomous weapons.