This paper speeds up diffusion language models (dLLMs) by changing the order in which they fill in missing words.
RecGPT‑V2 turns a recommender system into a smart team: a planner, several specialists, and a fair judge that all work together.