MultiMDM Lets Masked Diffusion Draft Plans First

Quanquan Gu, Associate Professor at UCLA and Pre-training & Scaling Co-Lead at ByteDance Seed, retweeted a post by Yinuo Ren. The post states that MultiMDM lets masked diffusion models draft before writing by allowing [MASK] to carry a rough plan for the eventual word. The description frames this as a capability in masked diffusion. No further details, results, or links appear in the packet.