AI still fails real science tasks at scale

A new benchmark using real scientific requests found that no frontier AI model consistently meets expert-acceptable quality.

Also today

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