Yair Schiff

🚨Coming to #ICML2026 🇰🇷

Last year, we introduced Block Diffusion LMs, adopted by dLLMs at Google, NVIDIA, Meta, etc.

🚀Today, we introduce Set Diffusion, a faster, more flexible alternative that combines AR and diffusion by interpolating generation orders.

🧵1/8

🚀 Introducing d2 — a principled and efficient RL framework for improving reasoning in diffusion language models (DLMs).

RL works well for autoregressive LLMs.
But for DLMs? It’s fundamentally harder.

We show how to do it right. 👇
📖 arxiv.org/abs/2509.21474
🌐 guanghanwang.com/d2
💻 github.com/kuleshov-group…

🧵1/12

🚨Announcing our #ICLR2025 Oral!

🔥Diffusion LMs are on the rise for parallel text generation! But unlike autoregressive LMs, they struggle with quality, fixed-length constraints & lack of KV caching.

🚀Introducing Block Diffusion—combining autoregressive and diffusion models for the best of both worlds!

👇1/7

We are excited to present ReMasking Diffusion Models (ReMDM), a simple and general framework for improved sampling with masked discrete diffusion models that can benefit from scaled inference-time compute.

Without any further training, ReMDM is able to enhance the sample quality of masked diffusion models on both unconditional and conditional generation by only changing the sampling algorithm.

🧵1/10