“The Flamingo model demonstrated an ability to converse about the subtle, joke-like context in an image of Barack Obama, a task previously highlighted by Andrej Karpathy as a challenge for computer vision systems.”