The ML community has been calling for a large scale climate dataset. In our recent Neurips 2023 publication, we introduce ClimateSet: A ➡️ large-scale ➡️ consistent ➡️ ML-accessible climate model dataset 🌍. 1/8 arxiv.org/abs/2311.03721 climateset.github.io

Apparently, you can run DOOM now on digestive system bacteria, technically making DOOM a shitty game. Congrats to MIT researchers who made this work! Source: docs.google.com/document/d/1SF…

Wuerstchen, a paper co-first-authored with dome | Outlier @pablopppgot Marc Aubreville Chris Pal into the top 5% (oral) at ICLR 2026 2024! Super proud of our work! Check out the paper here: surl.li/pjfiy

Würstchen received an Oral at ICLR WUHUUUUUUUUUUUUUUUUUUUUUU See you in May WUHUUUUUUUUUU openreview.net/forum?id=gU58d… Much love to Pablo Pernías Mats L. Richter Marc Aubreville Chris Pal ICLR 2026

What a ride! Congratulations to Pablo Pernías and dome | Outlier on your first paper ever - on ICLR of all places. Not many people can say this about themselves! Very proud 😊 and thanks a lot to Mats L. Richter and Chris Pal for this great collaboration!

It finally happened. We are releasing Stable Cascade (Würstchen v3) together with Stability AI ! And guess what? It‘s the best open-source text-to-image model now! You can find the blog post explaining everything here: stability.ai/news/introduci… 🧵 1/5

Congrats to my Coauthors dome | Outlier, Pablo Pernías, for going the extra-mile to scale Wuerstchen at Stability AI with #StableCascade. OpenSource, cheap training, fast inference, high-quality images, this is how you democratize AI! Check out our paper here, tinyurl.com/462ey8zk

@mk1stats made a colab version of Würstchen / Stable Cascade that can be run in the free version of Colab. Really cool! 😎 x.com/mk1stats/statu…

Interested in seamlessly updating your #LLM on new datasets to avoid wasting previous efforts & compute, all while maintaining performance on past data? Excited to present Simple and Scalable Strategies to Continually Pre-train Large Language Models! 🧵arxiv.org/abs/2403.08763 1/N

Here is the full paper of the continual pretraining project I have been working on last year. I encourage you to check it out if you pretrain LLMs (in particular, I recommend to start with takeaways in Section 2 and the Table of Contents at the start of the appendix).

🚀 Thrilled to introduce Recall to Imagine (R2I), the 1st model-based RL approach integrating SSMs to excel in memory-intensive domains. Not just setting new SOTA, but achieving superhuman results in complex memory tasks, while efficiently operating across diverse domains. 1/

Looking forward to being at ICLR 2026 in Vienna over the next week! Check out our oral presentation in Hall A 3 4:15 p.m. on Stable Cascade / Wuerstchen and meet us at the poster session right after our talk. You can find our paper here: openreview.net/forum?id=gU58d…

Our oral presentation at ICLR 2026 is happening at 3:45 in Hall A2! Discover how we achieved superhuman performance on memory tasks. Missed it? Catch us at a poster session in Hall B #183 from 4:30 to 6:30.

MediSyn: Text-Guided Diffusion Models for Broad Medical 2D and 3D Image Synthesis abs: arxiv.org/abs/2405.09806 Two instruction-tuned text-guided latent diffusion models, one for 2D medical images and one for 3D medical images. Trained on a dataset of 5.7 million 2D medical

Crowdstrike Analysis: It was a NULL pointer from the memory unsafe C++ language. Since I am a professional C++ programmer, let me decode this stack trace dump for you.

How do LLMs deal with misinformation? The answer is: not very well, but a natural resilience seems to emerge with larger models. Check out Mo Samsami 's latest work to know more!

Today, we’re thrilled to announce 3 major steps forward in bringing our vision of Agentic AI to life: 1️⃣ Runner H : Public Beta is now live! Imagine having your AI agent execute entire workflows across web apps, documents, spreadsheets, and more with a single prompt.

2️⃣ Holo-1: We are Open-Sourcing our Visual-Language Model powering Surfer H We’ve beefed up Surfer H’s web automation with Holo-1, our 3B & 7B-parameters Action Models. It now achieves industry-leading UI localization and navigation accuracy while staying compact and

A lot of work went into this! Proud of the team and what we've achieved.

🏄 Our Surfer-H agent, powered by our Holo1 model, hits 92.2% SOTA on WebVoyager! Achieves Pareto-optimal accuracy & cost-efficiency, outperforming GPT-4.1 and other foundation models at a fraction of the cost. hcompany.ai/surfer-h