Continuous Diffusion LMs went from niche to contender this year, but the recipe is shaky: why self-conditioning? How to embed tokens? 🤨 Our new work takes your one-hot flow from 4% to 50% on TinyGSM without self-conditioning, and explores the design space of CDLMs. 🔥🚀 (1/6)

Happy to share our new work: Distributional Functional Rewards (DFR)! 🎉 We improve diversity and fidelity in diffusion fine-tuning 🔧✨ Big thanks to my coauthors @JeongjaeLee0, Jong Chul Ye, @mmbronstein, @Yoshua_Bengio, @AlexanderTong7 and @danyalrehman17! 🙏 Thread below 👇

🔥 Diffusion LMs promise what autoregressive models can’t: fixing their mistakes. But new ones often change the noise and the sampler at once. Which matters? We tested it on the hypersphere. Noise: 0.8 points. Sampler: 25. We introduce Spherical Brownian Diffusion 🌐 👇🧵

🎉 Excited to bring our book The Principles of Diffusion Models to @NeurIPSConf 2026 as a tutorial—this time, all about diffusion LLMs! 🚀 🔹 How do ideas from continuous diffusion extend to language generation? 🔹 What are the latest advances and open frontiers in diffusion LLMs? From foundational principles to cutting-edge research. Stay tuned! 📆 Mon, Dec 7, 9 AM AEDT 📍 TikTok Ent Centre (Sydney, Australia) w/ organizers @ssahoo_ @mittu1204 @StefanoErmon

Excited to announce the follow-up to our NeurIPS 2026 Oral: Improved Training of Strong Stochastic Flow Maps! We improve the training of SSFMs by unlocking derivative-based losses! Joint work with @antoniofrancaib @rmincam @ChristianKoke @Niklas_TR and @AlexanderTong7

did a group presentation today with @3blue1brown's manim style transitions. Sharing the template for anyone to build upon Brought to life with @NarekMaloyan's manim-web + reveal.js + claude + codex view template: slides.ramith.fyi view repo: github.com/ramithuh/protein-…

Can a language model be based purely on self-conditioning? 🤔 Yes! Meet LAW 👋 (a.k.a. Loopholing All the Way!) 🧵1/8

1/9 Current optimizers are static during training. What if the optimizer could think and react while the network trains? We let an agent watch the training dynamics and rewrite the optimizer's source code as training goes: a reactive optimizer. In every language and image setting we tried that has no tuned recipe, it matches or beats the best recipe one can one-shot, and it guards against loss spikes and overfitting by itself. In the near future, optimization decisions may belong to agents, not humans. 📝 xingjianbai.com/reactive-opt… 💻 Try it on your task: github.com/xingjian-bai/opal 🧵

Can flow language models scale to billions of parameters and handle real generative tasks? Introducing 🌊 Lafa (Large flow language model), a flow LM scaled up to 8B that matches AR quality, with 🚀 2.5× AR throughput at 7B. Co-led with @ArnavChavan6 @UdbhavBamba. Joint work with @wognsfjq96 @hongseu33 @AdtRaghunathan @jw9730

⁉️ Is model distillation of frontier models like DeepSeek-R1 and Opus 4.6 traceable to how robust each model's distilled family is 🤔? We find evidence of a completely new phenomenon that we term "Vulnerability Monoculture," where distilled models inherit the security risks of their teacher 😱 🚨 Check out our new preprint where we expose the Systemic Adversarial Robustness (SAR) of open-source frontier models Punchline: We show that models distilled from DeepSeek-R1 and Opus 4.6 are systemically less robust! In some cases, 30- 40% collectively less robust 🚨🚨🚨🚨!!!! 📝Blog: leoelm.github.io/blogs/syste… 📄Paper: arxiv.org/abs/2610.27353 w/ @leo_elm @LSzpruch @matthew_wicker 🧵1/7

One fun part of this project was @nscorley @RafiBrent @maxkazman and me sitting late in the office, testing out some ideas in the recurrence space and a week later seeing a paper that echoes our findings. Go checkout @osclsd's recent work, super cool stuff!

We also found things that echo recent flow language model work. We replaced our long-term enemy, self conditioning (@osclsd), with recurrent updates, which others concurrently presented as looped flows (@nmboffi). These “thinking modes” boost performance further. 11/30

Super exciting to see the incredible advances of diffusion for bio too. Of course, with people like @DidiKieran, it's going to keep improving at incredible rates. Fun to see that self-conditioning is also falling in the protein space! Its days are numbered...

When training agents, you add a length penalty for exceeding the budget/effort level. The fun thing is, every lab (Deepseek, Kimi, ...) uses a different one. We compare which works best in our new paper. 📄 arxiv.org/abs/2610.28409 🧵1/9

🚨 Do you always have multiple rewards but don't know how to steer towards an expensive true reward? Don't fret, friends; we have the paper for you! 📣 Steered Not Stirred: Effective Reward Tilting using Surrogate Reward Mixing, led by the incredible @alec_sargood28 and team The Martini Algorithm 🍸 From 007: everyone now has a license to tilt 🤵 📜 Paper: arxiv.org/abs/2610.25502 🔗 Blog: alecsargood.github.io/martin…

🚨 New Paper Alert, 🎤 Accepted for an oral presentation at the NeurIPS 2026 STODY workshop! 📣 Introducing: Normalizing Flow Maps (NFM): Combining few-step transport from Flow maps with tractable likelihoods from Normalizing Flows. This is my first work with my first crop of new PhD students, @francescoshox and @tim_hitge, who genuinely knocked this project out of the park. 📄 Paper: arxiv.org/abs/2610.22711 📝 Blog: the-bose-lab.github.io/norma… 💻 Code: github.com/francescoshox/Nor… 🧵 1/8

More progress on CDLMs: self-conditioning has always been a bit of a crutch, DoubleCat/SemiCat seems to be a principled way to get rid of it🎉

oscar leave some work for the rest of us 😅 very cool idea and exciting to see more principled explorations of self conditioning!

Introducing RSI-Inference Bench from 🍎 Apple ML Research 🌐 rsi-inference.com 💻 github.com/apple-aiml-resear… Can AI agents improve the systems that run AI? We spent $100K API tokens, 50K GPU hours to find out. We built environments and benchmark from real-world #vLLM and #SGLang tasks, covering the full lifecycle of model inference engineering: 🚀 Day-0 Model Support ⚡ Inference Optimization 🔧 Bug Fix → Unsaturated: best agent at 26.7% pass@1 → Comprehensive coverage of the inference stack → Verifiable rewards: binary and continuous Watch the video to find out more!

Let's get it to sub 1 ;)

Introducing Langevin Flow Maps TLDR: Neural Network Potentials that learn their own stochastic pathwise solution for faster simulation. - Transferable across molecular systems. - Transition dynamics are preserved. - No trajectory data are required. Work together with @sgmccallum, supervised by @AlexanderTong7, and @JamesFosterBath. Arxiv: arxiv.org/abs/2610.05998 Data: huggingface.co/datasets/nikl… Code: github.com/sammccallum/lfm