Good morning 👋 You know that feeling when your carefully rehearsed presentation crashes at the worst moment, forcing you into awkward jokes about how technology hates you? Now imagine it happening live, streamed worldwide, with your reputation and stock price on the line. That was Zuck’s Wednesday when Meta’s new $799 AI glasses repeatedly glitched. At least the cringe reel is going viral, proving there’s no such thing as bad publicity.
🗝 OpenMind just dropped the 1st open-source OS for intelligent robots. Until now, building robots meant dealing with expensive software and locked-down hardware. OpenMind’s OM1 changes everything. It lets developers easily add advanced AI, vision, voice, and navigation to any robot – from humanoids to robot dogs – letting them think, learn, and work together. Robotics is having its Linux moment, hopefully without the cult following.
🍪 Holy crap. New details on China’s AI chips show jaw-dropping power. Huawei’s Atlas 950 SuperPod packs the punch of 1,000 Nvidia Blackwell GPUs, or about 2,000 current-gen H100 chips. But that’s just the appetizer. The upcoming Atlas 960 is nearly 4x stronger, and a planned “SuperCluster” (64 SuperPods strapped together) hits a mind-melting 250,000 Blackwell GPU equivalent. No wonder China dumped Nvidia.
👁️ A new protocol to fight AI deepfakes is a privacy nightmare. Backed by Big Tech like Google, Meta, and Adobe, C2PA tags digital content with metadata revealing who created it, when, and from exactly where. It’s already enabled by default on TikTok. Opt out, and your content looks suspicious. Opt in, and you’re giving Big Tech a detailed map of your identity. Your call: look shady or sacrifice privacy.
🤜🤛 Together with Sana
Work Faster and 10x smarter
Sana Agents give you expert no-code AI assistants in minutes, not months.
Why you need it: Teams waste too much time handling repetitive tasks. Sana Agents plug into your company’s tools – think Slack, Notion, Asana – to answer questions, handle admin, and scale productivity.
Use Sana Agents to offload repetitive tasks and free people to focus on work that actually matters.
🐳 China’s DeepSeek claims they trained R1 for just $294k. Pocket change. They say it took only 80 hours and 512 Nvidia H800 chips to train the advanced model. If that’s true, China is about to leave the US in the dust. But we call BS. Those chips alone cost around $18 million, not to mention salaries. Apparently that $294k covers only the electricity to run the cluster. Amazing what you can pull off when someone else picks up the tab.
💻 Someone just turned a disposable vape into a working web server. Programmer Bogdan Ionescu realized cheap, throwaway e-cigs have ARM microcontrollers powerful enough to host a website. The vape loads pages in just 160ms, although it’s currently offline. Right now, there’s literally a vape outworking your startup. What’s your excuse?
🔥 There’s an open-source tool that measures AI “stupidity” in real time. Every time a new model drops, benchmarks dominate headlines, and absolutely no one believes them. The “Stupid Meter” continuously tests popular models on coding, debugging, efficiency, and stability, tracking how many tries it takes to actually get the right answer. Finally, a benchmark that makes sense.
fun stats
raised by Colossal Biosciences to revive the Dodo bird 🦤 The Texas biotech startup is now valued at $10.3 billion, with plans to bring back the extinct woolly mammoth, Tasmanian tiger, and northern white rhino.