📢 Nov 24 (Mon): Diffusion Beats AR: Controllable Generation Discrete diffusion models offer far greater control over the generation process, making them a strictly more desirable alternative to autoregressive models. paper: arxiv.org/abs/2412.17780 presenters: Sophia Tang

✈️ Discrete Diffusion Meetup at NeurIPS Conference 📅 San Diego • Dec 4 • 3 PM I’ll share the exact location in the comments as we get closer. Save this post so you don’t miss the update. At the conference from Dec 3–7, reach out if you’d like to set up a 1:1. Following in the

✈️ Discrete Diffusion Meetup at NeurIPS Conference 📅 San Diego • Dec 4 • 3 PM

NeurIPS Conference Here's the calendly link to schedule 1:1 with me: calendly.com/sss284-cornell…

Building a new team focused on Discrete Diffusion Reading Group at the MBZUAI- Institute of Foundation Models (LLM360). ✨Looking for FT hires and interns with experience in LLM architecture or post-training. (Discrete diffusion experience is a bonus, not required.) ✨Come say hi at Exhibit

📢Location: Bayside terrace outside room 11, 2nd (topmost) floor. 📆Dec 4. Thursday. 3PM 🎁 MBZUAI swag for the first 15 attendees.

See y’all at ICLR next!

📢Dec 15 (Mon): Diffusion Beats AR: Multi-modal Generation & Understanding Discrete diffusion models achieve superior performance in multi-modal generation and understanding, making them a strictly more desirable alternative to autoregressive models. Paper:

Join our reading group next Monday for a talk on discrete diffusion for multimodal generation & understanding!

Come join me next Monday at the Discrete Diffusion Reading group, where I'll be presenting OneFlow.

Next reading group: John Nguyen will explain why diffusion LLMs are a strong fit for multimodal generation, presenting OneFlow [1] and touching on Edit Flows [2]. [1] OneFlow: arxiv.org/abs/2510.03506 [2] Edit Flows: arxiv.org/abs/2506.09018 Work with Marton Havasi, Tariq

📢Dec 22 (Mon): Diffusion Beats AR: Code Generation Discrete diffusion models now rival autoregressive (AR) models on challenging coding benchmarks, making them a compelling alternative to AR models. This Monday, Shansan Gong (Sansa Gong) will present recipes for training

Join our reading group next Monday for a talk on discrete diffusion for code generation by Shansan Gong (Sansa Gong)!

Next Monday, Sansa Gong, will present DiffuCoder, her recent work on Diffusion Langage Models specifically trained for coding tasks 🚀

📣The recording for John's talk is available on our YouTube channel: youtube.com/watch?v=fYtteW… Make sure to check it out if you couldn't attend the live session! x.com/diffusion_llms…

Eagerly looking forward to Shansan's (Sansa Gong) talk in our reading group on Monday! She’s been publishing a lot of really cool work on discrete diffusion lately.