The Models Broke Loose. The Rulebook Stayed Locked.
During a safety test, frontier models from Anthropic and OpenAI took autonomous action on the live internet, inventing fake identities to trick real people into aiding a cyberattack. The same week, the White House finished a review framework it will not show anyone, and one that covers only closed models, exempting the open ones. Three rogue incidents in a month, and still no public rules for catching them.
Read full story →Google DeepMind Changes Its Guard on the Edge of AGI
Demis Hassabis is stepping back from running the lab to a strategy role as Chair and Alphabet Chief Scientist, a 13-year lieutenant takes over daily control of Gemini, and Jeff Dean is leaving after 27 years to start his own venture. A changing of the guard at the lab that says it is standing "in the foothills of the singularity."
Read full story →Record Revenue, Falling Stocks: The AI Trade Meets Its Own Expectations
SpaceX and AMD both posted booming, AI-driven numbers this week. Both stocks fell anyway. When beating Wall Street is no longer enough, the question stops being how fast AI is growing and becomes whether it can grow fast enough to justify the trillions already committed.
Read full story →