Train complex first, simple follows

RL training on complex, multi-step tasks transfers to simpler ones better than training on parts first.

Also today

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