Unified AI Models Face Challenges in Multimodal Understanding | KnowAI Space