Tech »  How Idefics2 Answers the Unasked Questions in Vision-Language Modeling