Introducing Muse Spark, the first in the Muse family of models developed by Meta Superintelligence Labs. Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration. Muse Spark is available today at https://t.co/wHkMPH82ZH and the Meta AI app. We’re also making it available in private preview via API to select partners, and we hope to open-source future versions of the model. Learn more: https://t.co/PloE9q5x96

Releasing FLUX.2 Small Decoder: a faster, drop-in replacement for our standard decoder. → ~1.4x faster → Lower peak VRAM - decode larger images without running out of memory → Minimal quality loss → Works with FLUX.2 out of the box Especially impactful for real-time and larger resolutions pipelines.

Unreal Engine 5 is becoming the go-to platform for robotics teams, not just as a simulator but as a synthetic data factory powering the next generation of robotics systems🦾 From synthetic data to sim-to-real transfer, read our piece on how UE5 is quietly reshaping how robots learn to see, move, and act in the real world: https://t.co/Dc2VUvBs9q

We are giving away Safetensors to the @pytorch foundation (shepherded by the Linux Foundation) Our shared goal is to make the default serialization format for torch safe and performant. To unlock this, governance needs to be independent of @huggingface. Looking forward to more stakeholders contributing to Safetensors in the coming months 🔥

not surprised by any of this, headline or subheading https://t.co/ECcO5AMtid

Both societies exist simultaneously. really incredible to think about https://t.co/D2UHn3r03A

AI is set to reshape how cities function day to day. Samson Hu points to a future where infrastructure, services and urban systems are increasingly managed and optimized through AI. Cities are becoming systems of intelligence. How they operate will depend on how well that intelligence is integrated. https://t.co/N5rAQXqQ7E

Filmed at a real Primer school. Starring real Primer students. A 200 year vision for American Education: https://t.co/uTQDTJw633

"Anthropic, OpenAl and Google release their new models with high quality from day one then slowly nerf them until the next model, so when the next model hits, its perceived as a bigger jump than it actually is" sounds right what's happening https://t.co/D8FpbbYAwt

Introducing Avatar V. We’ve solved character consistency. Forever. Record yourself once for 15 seconds. From there, you can show up anywhere, in any look, and it still feels like you. Any photo becomes a video that looks, moves, and speaks like you, down to your mannerisms and quirks. This is the most advanced AI avatar model in the world. And we know that’s a big claim, so we brought the data to prove it. Thread below:

sam3.cpp - Meta's SAM 3 in pure C++ with @ggerganov's ggml - Supports SAM 3.1, 3, 2.1, 2 and EdgeTAM - FP16, 4-bit quant (EdgeTAM in 15 MB) - Apple Metal GPU, CUDA, CPU - Text-prompted: "peach" → every peach - Single-file C++14 Performance-wise: - 100ms object detection, segmentation - Video object segmentation @ 20FPS on M4 Pro with EdgeTAM https://t.co/XHC7ipyQtI

I’ve uploaded a new paper on arXiv (co-authored by @rasbt): MiCA Learns More Knowledge Than LoRA and Full Fine-Tuning In Parameter-Efficient Fine-Tuning, a key question may not just be how low-rank the update is, but *which* subspace we adapt. https://t.co/0Dc7F6rdGs

Exciting update to my 'now' page: I am taking some time to rest and cranking the explore-exploit dial all the way over to 'explore' for a bit 😁 https://t.co/U8GqPvPiJF

Claude Code Cheat Sheet Here is a list of top commands, shortcuts, and patterns you need to know when using Claude Code. We will update this as new updates for Claude Code come out. The idea is to curate practical tips and an easy way to navigate important functionalities and best practices. I have found this format more useful when learning what's new and teaching others about Claude Code. https://t.co/ScBEg1IwmY

“PyTorch is the most important open-source project at this moment in history.” https://t.co/lJv4TYnKy7

“PyTorch is the most important open-source project at this moment in history.” https://t.co/lJv4TYnKy7

And that’s a wrap. Thanks for an amazing first-ever #PyTorchCon Europe 2026! https://t.co/pRmqjLTGzn

+2 more

NEW paper from Google on multi-agent research agents. It's one of the first systems that handles end-to-end LaTeX generation, targeted literature reviews, and conceptual diagrams as a decoupled, standalone writer. Automated research frameworks can run experiments, but their writing modules remain the weakest link. Literature reviews are shallow, citations are sparse, and no system generates conceptual diagrams. This new research introduces a standalone writing framework that addresses all of this. PaperOrchestra is a multi-agent system that transforms unconstrained pre-writing materials, raw ideas, experimental logs, notes, into submission-ready LaTeX manuscripts. It uses specialized agents for deep literature synthesis, plot generation, conceptual diagram creation, and iterative refinement. The team also releases PaperWritingBench, the first standardized benchmark with reverse-engineered materials from 200 top-tier AI conference papers. Why does it matter? In side-by-side human evaluations, PaperOrchestra achieved absolute win rate margins of 50 to 68% in literature review quality and 14 to 38% in overall manuscript quality over autonomous baselines. Paper: https://t.co/HExJojfQPG Learn to build effective AI agents in our academy: https://t.co/LRnpZN7L4c

⏰ ICYMI — #PyTorchCon North America 2026 CFP is open! Got a session or poster idea? Submit for Oct 20–21 in San Jose. We're looking for work on training, inference, kernel engineering, responsible AI & more. 👉 Submit a session by June 7: https://t.co/Mz2gMtNnFc ⚡Submit a poster by July 26: https://t.co/kp9yuHyVtv

NEW paper: Memory Intelligence Agent (MIA) MIA boosts GPT-5.4 by up to 9% on LiveVQA. Quick summary: Most memory-augmented agents treat memory as a static retrieval problem. They store trajectories, retrieve similar ones, and hope for the best. But memory that doesn't evolve becomes stale, and storage costs grow without bound. This new framework combines a non-parametric Memory Manager for compressed trajectory storage, a parametric Planner trained via RL to produce search strategies, and an Executor that carries them out. The key innovation is bidirectional conversion between parametric and non-parametric memory, plus test-time learning that updates the Planner on-the-fly during inference. MIA boosts GPT-5.4 by up to 9% on LiveVQA. With a lightweight 7B Executor, it achieves 31% average improvement across eleven benchmarks, outperforming the much larger 32B model by 18%. Memory systems for agents need to evolve, not just accumulate. MIA's alternating RL training and bidirectional memory conversion show that treating memory as a living system, not a database, produces substantially better deep research agents. Paper: https://t.co/eSQa0URWCk Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX

first company i've seen with the stated goal of climbing the kardashev scale that actually might do it https://t.co/h07TvAMUj7

SKY FULL OF STARS: @elonmusk reaffirms his hope for humanity to reach new heights in outer space after NASA releases breathtaking new photo of the Milky Way galaxy captured by Artemis II astronauts. https://t.co/hEERowg4RF

PyTorch Ambassadors are making a big difference, spanning 21 countries and 5 continents. During the #PyTorchCon Europe 2026 Ambassador Meetup, ambassadors met with PyTorch Foundation leadership to share experiences https://t.co/KMFyvpRcKK

Glad to see that Mytos is putting back cybersecurity in the spotlight. Our main contribution to the topic has been the creation of Safetensors by @narsilou @huggingface in collaboration with @AiEleuther @StabilityAI. It has become the most popular way to share models safely and today @LysandreJik announced at #PyTorchCon that the project is joining the @PyTorch foundation to continue to scale (for example with torch core integration). Let's turn talks about cyber-security into action!

Google CEO Sundar Pichai on current frontier model's ability to break the security of almost all current software. "These models are definitely, like really gonna break pretty much all software out there, maybe already, we don't know." https://t.co/ezjPcSehiG

This is a great tutorial (credits @itsclelia + @lancedb) on how to build a practical retrieval pipeline that integrates directly with your agent harness. 1. Ingest a massive pile of docs with liteparse. 2. Store data in a vector db (despite my memes to the contrary, you will need some database for larger scale retrieval). 3. Pair with image screenshotting tools that allow the agent to "dive deeper" into data. When you pair this with the Claude Agent SDK / Claude Code, the agent will do some initial retrieval pass to pull the relevant doc, and then use screenshotting/VLM-enabled capabilities to do deeper analysis. Blog: https://t.co/mbdy4I5oAh

Safetensors is joining the @PyTorch Foundation as a foundation-hosted project It has always been open-source. Going forward, strategy and direction is now open Now, safetensors has a clear path towards integration in torch core ML is big, so is making it safe. Come contribute! https://t.co/yLQFgxNJDs

Your brand is an afterthought for out-of-the-box AI tools. Of course it is. They weren’t built for you, they were built for everyone. You deserve a creative production platform that puts your brand first, exactly how you envisioned. Introducing Brand Studio by Stability AI, the creative production platform powered by your brand. Get started here: https://t.co/WUWML6YJ28

Our team (@kstonekuan , @Artylo99) came in 2nd at the OpenAI x January Capital hackathon! We're grateful for the organizers' generosity and support. Excited to keep building alongside the community. https://t.co/hIT0jPoya0

We released Gemma 4 last week, and seeing the community's response has been amazing! 🚀 Honored to lead the vision efforts in which we made huge performance leaps from Gemma 3, I wanted to help you make the most of the new capabilities. Deep dive 🧵 https://t.co/nP6ogGgghf

ETERNO is supercharging healthcare by transforming pen and paper practices into cloud-based clinics that support patients and clinical teams. Built on AWS, its platform streamlines operations, strengthens data protection, and provides a foundation for ETERNO to scale in Europe. https://t.co/IAgkYi0WjL

People of pi. BIG NEWS. I've sold out. Let me know how you feel about this in the comments below. https://t.co/dy1ffkhMRf