Personal AI should run on your personal devices. So, we built OpenJarvis: a personal AI that lives, learns, and works on-device. Try it today and top the OpenJarvis Leaderboard for a chance to win a Mac Mini! Collab w/ @Avanika15, John Hennessy, @HazyResearch, and @Azaliamirh. Details in thread.

@lennysan @TheRialMichelle Really enjoyed this one. It was reassuring having someone to else acknowledge the tradeoffs especially “the treadmill” which is so true (but a good tradeoff 😃)

@HanchungLee Why

@itamar_mar Wow. Interesting results. Great to see attention to recall which matters a lot in this use case. And even more impressive how the specialized agents catch the remaining ground truth. https://t.co/WcVSS59rO9

We're sharing a new method for scoring models on agentic coding tasks. Here's how models in Cursor compare on intelligence and efficiency: https://t.co/VItnifMh55

About to try to upload 100TB to try out HF Buckets... 👀

Paper drop, 3 years in the making. Ever felt the model "helped" but somehow made things worse? Now we can measure it: AI proactivity imposes cognitive load that degrades your work - and once the model derails, it doesn't recover. You do. 🧵 https://t.co/zWpYL6vCsg

More evidence that we have to figure out how to improve the way humans and AIs work together, or we humans will end up overwhelmed.

I graphed a lot of the most critical and high-quality benchmarks, including some very new ones that no model has optimized for, and the curves have all had a very similar shape over the past year. https://t.co/C3xuZgeOJl

MXFP8 training for MoEs on GB200s enables a 1.3x speedup with equivalent convergence versus BF16. This #PyTorch update via TorchAO and TorchTitan on Crusoe Cloud details gains from dynamically quantized grouped GEMMs. #AI #OpenSource 🔗 https://t.co/CifP9fQBEC ✍️ @vega_myhre, @_xmfan, @drisspg, Chinmay Baikar

An internal AI incident at Amazon reportedly triggered a mandatory meeting over its “high blast radius.” As AI systems become deeply embedded in operations, a single failure can cascade across products, customers and infrastructure. Scale is AI’s superpower, but it also magnifies risk. @ElonMusk’s warning to “proceed with caution” raises a bigger question: are companies moving faster with AI than their safety systems can handle? https://t.co/6Q9PMnaRFQ @fortunemagazine @sashrogel

Introducing Mixedbread Wholembed v3, our new SOTA retrieval model across all modalities and 100+ languages. Wholembed v3 brings best-in-class search to text, audio, images, PDFs, videos... You can now get the best retrieval performance on your data, no matter its format. https://t.co/PYT3Ryerxm

I rare interview with me! By my wife! My brilliant wife (@TheRialMichelle) is about to come out with her first children's book, Charts for Babies, and I thought what a fun excuse to have her come on the podcast turns the tables on me. She asked things no one else would think to ask, and many things I've never shared publicly. We chat about the specific moments that pushed me to start the newsletter, how I think about quality, what stresses me out most, the invisible treadmill built into creator businesses, and how a psychedelic experience gave me the confidence to do this work. This was so fun, and so special, and I hope you like it. Listen now 👇 https://t.co/8JhYLb52DR

I rare interview with me! By my wife! My brilliant wife (@TheRialMichelle) is about to come out with her first children's book, Charts for Babies, and I thought what a fun excuse to have her come on the podcast turns the tables on me. She asked things no one else would think to ask, and many things I've never shared publicly. We chat about the specific moments that pushed me to start the newsletter, how I think about quality, what stresses me out most, the invisible treadmill built into creator businesses, and how a psychedelic experience gave me the confidence to do this work. This was so fun, and so special, and I hope you like it. Listen now 👇 https://t.co/8JhYLb52DR

Microsoft's VS Code product lead on how GitHub Copilot and custom AI agents helped the team move from monthly to weekly releases. https://t.co/q1QXM99ke1

You may have noticed that @code has moved to weekly releases - @pierceboggan sat down with @thenewstack to share how the team's approach to building with AI helped enable this!

@tunguz @basche42 Yes

🚨BREAKING: PENTAGON THINKS CLAUDE HAS BECOME SENTIENT AND MAY SOON TAKE OVER >claude has a soul >has a constitution >that’s NOT the U.S. Constitution >the other day claude was anxious >anthropic believes claude has a 20% chance of being sentient right now >has its own ability to make decisions “So does the Department of War want something like that in their supply chain?” >embedded across defense contractor systems via palantir >could hallucinate and poison the system >legitimate threat to national security >weapons, aircraft, war fighter protection >all compromised Pentagon CTO: “that’s really where the supply chain risk came from”

Wild. If any of this is true (and most if it isn’t*) it’s true of all LLMs and not just Anthropic’s models. There is no technical difference between models that would make one more of a supply risk than another. *mimicking text doesn’t make an AI conscious or anxious etc. but again all models do such things.

The people making decisions about AI in the US really don’t seem to understand how the generative AI models work or what intelligence is, or how to evaluate it.

ANNOUNCING https://t.co/iMvfCQ955F Upload a short video directly from your phone. Extract grounded 3D spatial intelligence. One of the biggest bottlenecks to open-world robotics is contextually useful memory. At Dwell Bot we're building it.

ANNOUNCING https://t.co/iMvfCQ955F Upload a short video directly from your phone. Extract grounded 3D spatial intelligence. One of the biggest bottlenecks to open-world robotics is contextually useful memory. At Dwell Bot we're building it.

Ukraine opens battlefield data to train AI models for autonomous systems — first in the world 🤘 Millions of annotated combat frames from thousands of missions are now available for partners to train AI. This creates a new win-win cooperation model: partners refine their technologies with unique data, while Ukraine scales frontline capabilities. Better data → smarter drones → faster decisions on the battlefield. The future of war is autonomous.

Your videos can go further now. We’re introducing new Video API capabilities, powered by Sora 2: • Custom characters and objects • 16:9 and 9:16 exports • Clips up to 20 seconds • Video continuation to extend scenes • Batch jobs for video generation https://t.co/CHpRpSi7TV