Evangelos Kazakos

📣 Announcing the release of the 🕊️ Annotated Encyclical 🕊️ from the ethics & society folks at @hf.co. Includes citations to relevant academic work. Very much a WIP. Please add work we haven't added yet! huggingface.co/spaces/socie...

The Annotated Encyclical - a Hugging Face Space by society-ethics huggingface.co

BAAI just released the Orca paper 🔥 ( weights coming soon ) huggingface.co/papers/2606.... A Multimodal Latent World Model: it learns the world itself first, and text/images/actions are just different ways to read it out 💡

Paper page - Orca: The World is in Your Mind huggingface.co

🔥 Our latest work is state of the art 3D pose estimation, with video analysis from a single camera! 🎥 Just train with 3D data, and then only 1 camera is needed! 🖤 Presented this week at #CVPR2026 by lead author Ti Wang! Please repost🩵🙏🏼 xiu-cs.github.io/FMPose3D/ arxiv.org/abs…

FMPose3D: monocular 3D pose estimation via flow matching arxiv.org

Coming back from #CVPR2026 in Denver after spending almost a week. I gave a short oral talk this year at the @how_cvpr2026 on "A Case Study on Hidden Bias in Vision-Language Model Activations." Feel free to check out the paper recently published with Mahzarin Banaji!! arxiv.or…

Vision-Language Models Suppress Female Representations Under Ambiguous Input arxiv.org

Introducing RoMo, a dataset of 820K high-quality 3D human motion clips, featuring a detailed semantic taxonomy. This innovation bridges the motion data gap, allowing generative models to achieve high fidelity and nuanced understanding of complex prompts. https://arxiv.org/abs/…

RoMo: A Large-Scale, Richly Organized Dataset and Semantic Taxonomy for Human Motion Generation arxiv.org

A study shows that vision-language models rely on visual encoders for spatial reasoning in tasks like image captioning. By enhancing vision-derived ordering signals, researchers achieved improvements in model performance, paving the way for more efficient AI. https://arxiv.org…

The Dual Mechanisms of Spatial Variable Binding in Vision-Language Models arxiv.org

Research exposes flaws in Vision-Language Models' chronological reasoning, revealing their reliance on superficial cues like grayscale images over accurate historical semantics. New datasets seek to guide the development of more logically driven multimodal AI. https://arxiv.or…

Seeing Time: Benchmarking Chronological Reasoning and Shortcut Biases in Vision-Language Models arxiv.org

social media platforms that operate to maximize engagement, nazi ideology built on hate and rage, weak regulation and enforcement... democracy is dying before our eyes and big tech is fuel powering it www.irishtimes.com/world/europe...

AI software that generates ‘rage bait’ developed by Germany’s far-right AfD – The Irish Times irishtimes.com

ByteDance dropped Lance👀 huggingface.co/bytedance-re... This 3B model can generate images + edit images + generate videos + edit videos, and understand both images/videos. It's trained from scratch on only 128 A100s, and beats several 7B+ models on GenEval and VBench!

bytedance-research/Lance · Hugging Face huggingface.co

Not so long ago we expanded Dynaword to Swedish, Norwegian and Icelandic Now have a new addition - Faroese 🇫🇴 If you are working on Faroese NLP or know someone who works on it, we would greatly appreciate any advice regarding what datasets we should add -> huggingface.co/datas…

danish-foundation-models/faroese-dynaword · Datasets at Hugging Face huggingface.co

Meta Lays Off 8,000 Employees, as A.I. Casualties Mount www.nytimes.com/2026/05/19/t...

nytimes.com