no pretrained encoder, no complex tricks. LeWorldModel shows how JEPA-based World Models can be trained end-to-end from raw pixels with just 2 loss terms ~15M params, single GPU, and ~48× faster planning than foundation-model world models. https://t.co/0bUVowVbo6

@ExaltedFoks @leadrnm https://t.co/IfTJYuUv7m

JEPA are finally easy to train end-to-end without any tricks! Excited to introduce LeWorldModel: a stable, end-to-end JEPA that learns world models directly from pixels, no heuristics. 15M params, 1 GPU, and full planning <1 second. 📑: https://t.co/cpTzgvbTS0 https://t.co/Z2De9ASzcW

As AI systems move from training to real-world deployment, 2 forces are rapidly reshaping the landscape: the explosive growth of AI inference & the emergence of agentic AI. Don’t miss the KubeCon + CloudNativeCon Europe 2026 fireside chat tomorrow AM that will explore how the rapidly expanding inference market is driving new infrastructure needs with @addvin - @RedHat, @linsun_unc - @soloio_inc, @jbryce - @linuxfoundation, @sparkycollier - @PyTorch 📅 Tues, March 24 ⏰ 09:56 CET Session details: https://t.co/Sn8iJfwZ3L #OpenSource #GenerativeAI #KubeCon #CloudNativeCon #AIInference #AgenticAI #AI

+1 more

Remix spins up parallel AI agents that turn your existing data into ready-to-post content — images, videos, articles, tweets, even apps. No prompting. No editing. Just publish. Download now: https://t.co/S7D9QUI2Sg Congrats on the launch, @samrkaplan! https://t.co/gD42unu9NT https://t.co/3kdEb2PU2f

Alibaba released LumosX on Hugging Face An ICLR 2026 framework that relates any identities with their attributes for personalized multi-subject video generation using relational attention mechanisms. https://t.co/w6eLDbU5HB

Uni-1 is here! A new kind of model that thinks and generates pixels simultaneously. Less artificial. More intelligent. https://t.co/2p8kSq4Jtf

I did not call Moby “little idiot” that is his twitter/x name/handle. I would never insult someone over their stature https://t.co/qDnoomDs9v

When things become too easy, something else can get lost. If AI removes friction from thinking, problem-solving and effort, it may also reduce the sense of progress, learning and satisfaction that comes from doing the work ourselves. Convenience has a hidden trade-off. The risk is not only dependency, but a gradual weakening of the skills that effort once built. https://t.co/FWMkU6uNmy https://t.co/FWMkU6uNmy @IEEESpectrum @vanessabramirez

Im crying Trump's Iran strategy is basically this https://t.co/pFxxNhMnZc

Im crying Trump's Iran strategy is basically this https://t.co/pFxxNhMnZc

NVIDIA's Kimodo is the release of the week 🔥 Prompt the timeline whatever your want like: "a person walks forward" → "a person starts jumping", hit Generate, and watch a 3D character do it in seconds (700hrs of pro mocap training. Works on human + robot skeletons. Super fast + free to use on HF)

HopChain Multi-Hop Data Synthesis for Generalizable Vision-Language Reasoning paper: https://t.co/20RhzWmDJM https://t.co/IZe1ZLXJhZ

+1 more

Astrolabe Steering Forward-Process Reinforcement Learning for Distilled Autoregressive Video Models paper: https://t.co/xVgmpyTRyO https://t.co/rh5mTdh7Vr

chicago. delivery robot drove through the glass of a bus shelter. this country kicks so much ass. https://t.co/eLbCmgoZqO

chicago. delivery robot drove through the glass of a bus shelter. this country kicks so much ass. https://t.co/eLbCmgoZqO

I like how every other platform has perfected targeted advertising but twitter is still serving up stuff like this https://t.co/H6g5vPRXea

Interesting finding in this paper showing that, for product development ideas, AIs consistently rank above humans (well, humans on Prolific) & larger and more recent models are more creative than previous ones. (It also tries a creativity intervention that doesn’t work on LLMs) https://t.co/7nK1RJAhyS

@ItzSuds @shit_queen @somewheresy 1200 from jfk to sfo https://t.co/CDE6GuViQC

We want to make @huggingface buckets the S3 for agents! Let us know if you can use something like that and happy to work with you on your usecase https://t.co/HExdOvLica

Did you like @victormustar’s DLSS 5 Anything @Gradio Space? 🧨 Now, HF Pro users can run batch image enhancement using @HuggingFace Jobs 🤗 Try it: https://t.co/Mxc4TmPGhA 🪭

Got to meet the wonderful Chip Huyen @chipro She’s so nice and smart!! https://t.co/IKqfmUeVMJ

Got to meet the wonderful Chip Huyen @chipro She’s so nice and smart!! https://t.co/IKqfmUeVMJ

my kid is non-verbal and autistic. the tools that exist for him cost thousands, lock you in and haven't kept up with AI. so i built an AI OS jr for him with his therapists. it's free. it learns how he communicates. and now its for all children who need it https://t.co/B4CSjMsxer

If you've ever worked in or around legal, you know that discovery is where document parsing really gets stress-tested. Low-resolution scans. Black and white images. Handwritten annotations. Charts buried in reports. Files that are technically PDFs but practically unreadable. And hundreds of thousands of them. Traditional OCR tools struggle with degraded scans, and anything visual (photographs, slide decks, tables) falls through the cracks entirely. That means your search index is noisy, your recall suffers, and relevant documents go unfound. This blog by @tuanacelik walks through how to set up LlamaParse for a legal discovery use-case: handling difficult scans with vision models, surfacing image and chart content, and using custom parsing instructions to guide output for predictable document patterns. The quality of everything downstream depends on what happened at ingestion. Worth getting right. Read the full blog here: https://t.co/MkUWjaJzSm

OpenClaw 2026.3.22 🦞 🏪 ClawHub plugin marketplace 🤖 MiniMax M2.7, GPT-5.4-mini/nano + per-agent reasoning 💬 /btw side questions 🏖️ OpenShell + SSH sandboxes 🌐 Exa, Tavily, Firecrawl search This release is so big it needs its own table of contents. https://t.co/XvRbXEduGC

📢WorldAgents: 3D worlds only from 2D image models - without any training! We propose an agentic approach with a Director (VLM) to plan the scene, a Generator (Flux or NanoBanana) for new views, and a Verifier (VLM) for selection / 3D consistency. -> High-fidelity 3D worlds from a single text prompt. What's remarkable: our agents find consistent views from 2D image models to obtain 3D-consistent worlds; this shows that image models contain world priors - agents just need to find them! https://t.co/6NC7zIEn4n https://t.co/vTO3sLFLFw Great work by @ErkocZiya @angelaqdai

agriculture changes forever starting today. introducing FAMA, our gen 1 autonomous robot that plants seeds, fertilizes, and weeds farms. fully self-driving, no GPS. we had no hardware background when we started. but we pulled through. more robots for farming incoming 🚀 https://t.co/HKyDFAayCx

JFK rn https://t.co/kaHMMRIaCD

Today we're launching Latent-Y: the world's first autonomous agent for drug design, lab-validated end to end. Give it a research goal. Latent-Y reasons, designs, iterates, and delivers lab-ready antibodies, autonomously or collaboratively, with the biological reasoning of a PhD protein design expert. Technical report: https://t.co/E7IHfkvvD3 Blog post: https://t.co/GfJAfzj0Qx Apply for access: https://t.co/E0SR9znZiP

If you're building anything in AI, the best skill you need to be using right now is hugging-face-paper-pages Whatever problem you're facing, someone has probably already published a paper about it. HF's Papers API gives a hybrid semantic search over AI papers. I wrote an internal skill, context-research, that orchestrates the HF Papers API into a research pipeline. It runs five parallel searches with keyword variants, triages by relevance and recency, fetches full paper content as markdown, then reads the actual methodology and results sections. The skill also chains into a deep research API that crawls the broader web to complement the academic findings. The gap between "a paper was published" and "a practitioner applies the insight" is shrinking, and I think this is a practical way to provide relevant context to coding agents. So you should write a skill on top of the HF Paper skill that teaches the model how to think about research, not just what to search for.

Underneath the Composer controversy, it's pretty wild to see the progress on Terminal-Bench 2.0. Open models are now stacking above 50% — 8 months ago, the best was under 28%. Made a @huggingface Space to see how models have progressed on important benchmarks. https://t.co/nELCt0eY4Z