AI safety needs better power metrics

A new framework reveals why common measures of human control in AI systems can mislead, and proposes a better floor for monitoring user power.

Also today

Today's top five AI stories, one line each, at 7am Eastern. Free. One-click unsubscribe.