RL agents win by arriving, not deciding

A new evaluation method splits RL agent gains into two parts: reaching good states versus solving from them.

Also today

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