Model identifies instances when autonomous systems have learned from examples that may cause dangerous errors in the real world. https://t.co/aNJYwoGs4L pic.twitter.com/HGOPiRqk7Q