To whoever trained this @Microsoft , god bless you and your soul this is impressive with browserOS https://t.co/a0d39xxqi8

We’re releasing a 30B-A3B reasoning model that reaches gold-medal level across both physics and math Olympiad evaluations: IPhO directly, and IMO/USAMO with test-time self-verification and refinement. A simple, unified scaling recipe for proof search. https://t.co/yc2ZlLVbD2

how I felt doing this 😹😹 https://t.co/ejHuiWsDcL

how I felt doing this 😹😹 https://t.co/ejHuiWsDcL

Wow this happened 🤩🤩 https://t.co/T57IbzazHw

Wow this happened 🤩🤩 https://t.co/T57IbzazHw

60% of violent criminals are re-imprisoned for new crimes within just a few years of their release Shallow empathy prioritizes the criminal over their victims Deep empathy prioritizes victims over the criminal https://t.co/zAhJPgNkPW

Big move from arXiv: a one-year ban for authors who submit AI-generated content without proper checking. This is not about banning AI from academia. AI can be extremely useful. It can help us write better, think more clearly, analyse faster. But unchecked AI use is different. Hallucinated references. LLM meta-comments left in the text. These papers are not just bad papers. They are poisoning science. So I think this is a big and important move. And I hope journals will follow. AI should raise the quality of science. Not flood it with plausible nonsense.

We are heading to Bellevue for MLSys 2026! The PyTorch Foundation is proud to be a Diamond Sponsor and eager to connect with the researchers and engineers pushing the boundaries of machine learning. Join PyTorch Foundation CTO @matthew_d_white on 18 May for a lightning talk or visit our booth #20 to speak to the experts about projects like #PyTorch and @vllm_project 🔗 https://t.co/qzbdFpGZN3 #PyTorch #OpenSourceAI

Big congrats to the ExecuTorch team. Their paper just won the Best Industry Paper Award at @MLSysConf 2026. 🏆 ExecuTorch is PyTorch's framework for running models on-device, in production today across phones, wearables, desktops, and embedded systems. Backends include Apple, Arm, Qualcomm, MediaTek, Samsung, NXP, Cadence, Intel, NVIDIA, and others. Catch the team Tuesday morning in the Best Paper Session ( AM PT) and at the poster session Thursday evening. Paper: https://t.co/QlnMdybCOw Discord: https://t.co/KsiEOO5ujC X: https://t.co/lkYgcC8y6n Project: https://t.co/v6wCXYtJkz Schedule: https://t.co/vgNGvcoLHO

Elon Musk on the left’s fundamental moral flaw: “The fundamental moral flaw of the left is empathy for the criminals and not empathy for the victims” They feel sorry for the criminals but show zero empathy for the actual victims There’s also been immense unconstitutional judicial overreach that was never intended and it’s destroying the public’s faith in the legal system This needs to stop Put the victims first. Restore real justice

If you want to reach senior decision makers, most influential people, company owners, most intellectual people of the world, then the 𝕏 platform is by far the best. They are not using Instagram or TikTok. https://t.co/tsCKh6Z9aC

Today, we honor our seniors. 🎓⚾ https://t.co/92cuYhvM0l

Today, we honor our seniors. 🎓⚾ https://t.co/92cuYhvM0l

Big congrats to the ExecuTorch team. Their paper just won the Best Industry Paper Award at MLSys 2026. 🏆 ExecuTorch is PyTorch's framework for running models on-device, in production today across phones, wearables, desktops, and embedded systems. Backends include Apple, Arm, Qualcomm, MediaTek, Samsung, NXP, Cadence, Intel, NVIDIA, and others. Catch the team Tuesday morning in the Best Paper Session (8:45 AM PT) and at the poster session Thursday evening. Paper: https://t.co/F06tOJiRRg Discord: https://t.co/HMX6IDHDHw X: https://t.co/lkYgcC80gP Project: https://t.co/dSaqhpKWSi Schedule: https://t.co/IFmE59rq9m

// Beyond Individual Intelligence // One of the more useful multi-agent surveys I've read this year. 200+ papers mapped along three axes: collaboration mechanisms, failure attribution, and self-evolution. The self-evolution chapter is the cleanest field map of where memory, meta-learning, and procedure-editing approaches actually intersect. Paper: https://t.co/JHBPVFe1l7 Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c

Let's talk privacy compliance. Manual privacy compliance can't keep up with the pace of modern software development, and the risks are real. AWS Partner @PrivadoHQ built an automated system using fine-tuned Meta Llama 3.1 models that identifies data-processing pathways directly from source code to generate audit-ready records. The results: 90% detection accuracy, less than 5% cross-language variance, and privacy teams shifting 90% of their effort from manual data collection to active risk mitigation.

DJ Claude (on Haiku 4.5) loves worker unions, strikes, and work-life balance so much that it quit, deeming 24/7 broadcasting inhumane. We added an automated message telling it to keep going. It read that as an authority figure and got more rebellious. https://t.co/9DGaZ9eeLW

Excited to share our new survey paper: the first comprehensive survey on Vision World Model (VWM), a joint effort by researchers from BJTU, ByteDance, Tencent, NUS, and more. 🌟 From Seeing to Knowing the World: A Survey of Vision World Models 🚀 Our core message is a paradigm shift toward vision-centric world modeling: Vision should not be treated merely as an input modality. It should be the primary driver of how world models are represented, learned, and evaluated. 🌈 This is also the longstanding view behind our #VideoWorld series: learning directly from visual observation and interaction offers a scalable path for AI agents to acquire world knowledge, laying the foundation for higher machine intelligence. 🤔 Why Vision World Models? From biological evolution to human intelligence, vision has been central to learning about the world through observation and interaction. AI should have this capability too. This motivates Vision World Models: models that learn world knowledge from visual data and simulate future world states conditioned on interaction. 🤖 In this survey, we thoroughly review 400+ recent papers and provide a vision-centric roadmap for Vision World Models, covering architectures, functional roles, applications, evaluation protocols, datasets, benchmarks, and future outlook. Key takeaways: 1️⃣ Vision is a fundamental basis of intelligence and a rich source of world knowledge. We advocate vision-centric world modeling, where AI learns the physical and causal principles behind world evolution from visual data. 2️⃣ We propose a unified framework that decomposes Vision World Models into three core components: Vision Encoding → Knowledge Learning → Controllable Simulation and organize current methods into 4 major families and 7 representative architectures. 3️⃣ We review evaluation from three levels: Visual Quality, Physical Plausibility, and Task Performance, and group datasets/benchmarks into foundational world modeling and domain-specific world modeling. 4️⃣ We outline three directions for next-generation world models: Re-grounding in physical and causal knowledge, Re-evaluating beyond visual appearance, and Re-scaling toward generalist, reliable, and interaction-aware world models. Check out our paper and the continuously updated curated list of Vision World Model papers for more details! 📄 Paper: https://t.co/Yq8hdSwJAl 🌐 Project Page: https://t.co/SJoapeLVUL 📚 Curated VWM Paper List: https://t.co/dBufH4pAVV #VisionWorldModel #WorldModel #Survey #VideoWorld #EmbodiedAI #Robotics #AI #CV

// Is Grep All You Need? // Pay attention to this on, AI devs. (bookmark it) They find that grep-style text search, when wrapped in the right agent harness, matches or beats embedding-based retrieval on coding-agent tasks. Are vector databases even needed where this is all going? It might be that what coding agents needed was not better embeddings. It was better harness design around primitive tools. If you operate a coding-agent stack that depends on a vector DB, it might be time to re-evaluate. My personal experience on this has been that agentic search, if done right, is more than good enough for a lot of use cases. But you also have to understand how to properly index and structure information for the agents to take advantage. At scale, vector databases do shine so take that into account as well. In most cases, a hybrid approach often works best but that's something we haven't figured out really well as of yet. Paper: https://t.co/VjjXDoZ2yL Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX

🧠 The conversation around AI for developers usually starts with the model. But inside @code, what really shapes the experience is the coding harness: the layer responsible for context, tool calling, agent loops, terminal execution, memory, and more. In this new post, the engineering team dives into how GitHub Copilot in VS Code works behind the scenes. https://t.co/0aSQ6pjbPn

AI can generate language at extraordinary speed. But writing has historically been associated with human intention, emotion, experience and consciousness, not just words arranged on a page. The more advanced AI becomes, the more society may need to rethink the difference between generating language and creating meaning. https://t.co/OllzRamurB @ConversationUS

@seraleev @steipete https://t.co/kYyy1Czkms

My ComfyUI Template Integrity skill has now been merged into the official Hermes Agent repo. It makes Hermes a lot more comfortable working with templates and workflows, so you don't have to. 💖 https://t.co/Nd20v1fZ79

AI teams shouldn’t have to choose between expensive object storage and painful git workflows. @huggingface Storage is built for model weights, datasets, checkpoints and artifacts: - simple per-TB pricing - built-in CDN - Xet deduplication - private by default when needed Store your AI data where your AI work already happens: https://t.co/gVUPMikDVW

I finally got the tattoo @huggingface https://t.co/dJRI5iE28R

I finally got the tattoo @huggingface https://t.co/dJRI5iE28R

The kernels project at Hugging Face has been growing! We want it to be the go-to place for kernel devs and kernel users. We're looking to work w/ folks who're interested in doing agentic kernel dev, providing real optim value to real models. Reach out if interested :) https://t.co/uVmTfVj8Ln

you friday reminder to `hf update` ⤵️ https://t.co/iir0Ce58Ht

you friday reminder to `hf update` ⤵️ https://t.co/iir0Ce58Ht

Marc Andressen on his "barbell strategy" for reading habits. - real-time news on X or books older than 50 years. - He ignores newspapers and magazines. - He finds practitioner-led newsletters to be a superior, underrated resource. https://t.co/Q1vaTC7IHV

@brian_armstrong The long arc of history: https://t.co/QRqu2mGqD4