Reasoning: Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training | Foresight | ScienceToStartup