Vision training shapes how AI thinks in text

Multimodal training reshapes how language models reason about spatial tasks, even when no image is present.

Also today

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