DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models - Explained Simply | ArXiv Explained