GPT-5.2 is still struggling with video analysis. I threw an Ilya Sutskever clip at it, and Gemini nailed the full transcript + context, while GPT choked. The multimodal gap is WILD.