Visual Generation Unlocks Human-Like Reasoning through Multimodal World Models | ScienceToStartup | ScienceToStartup