I have a hard time believing that this is some sort of emergent behavior and not the result of intentional training to scale AI research.