Computer vision has remained at the task-specific model stage; GenCeption wants to move past it.