Researchers at @BroadInstitute, @UniofExeter, and beyond are already using AlphaGenome Atlas to better identify potential disease-causing DNA variants and interpret their role. 🧵
Introducing OUI-1: the first open-weights model for Generative UI
71.7% on Generative UI Bench at 4B params. Beats Gemma 4 31B with 8× fewer active params, and scores 5.5× the base DiffusionGemma it was fine-tuned from.
Methodology, weights, and full benchmark results in the blog 👇
#ULTIMAHORA Fuertes declaraciones de China a EEUU:
Si Estados Unidos aún tuviera confianza en su propia competitividad, no necesitaría convertir cada proyecto civil chino en una emergencia de seguridad nacional.
Un puerto se convierte en una amenaza.
Un ferrocarril se convierte en una amenaza.
Una red de telecomunicaciones se convierte en una amenaza.
Incluso un cable submarino se convierte en un arma.
China construye infraestructura.
Washington construye pánico a su alrededor.
América ya no pregunta si un proyecto chino funciona.
Pregunta qué tan rápido puede ser destruido antes de que el mundo note que funciona mejor.
Cuando un imperio ya no puede liderar, intenta incapacitar a aquellos que sí pueden.
¿Cómo puede contestar la Casa Blanca? Seguramente con una guerra
#UrgenteCOMPARTAN #RTMasivo 👇🏼👇🏼👇🏼👇🏼👇🏼👇🏼👇🏼
Free and open-source Rufus-style USB creator for Mac that builds legacy macOS bootable drives on Apple Silicon, plus Windows and Linux installers. No Terminal required
🚨 ÚLTIMA HORA: Microsoft acaba de presentar el Surface Laptop Ultra, el primer PC con Windows diseñado para competir directamente con el MacBook Pro.
Ultrafino, incorpora un chip ARM de NVIDIA con RTX Spark y hasta 128 GB de memoria unificada.
El dispositivo está pensado para desarrolladores que quieren ejecutar agentes de IA de forma local, sin depender de la nube.
Su lanzamiento está previsto para el segundo semestre del año.
Durante años, Apple dominó este mercado sin competencia.
Eso acaba de cambiar.
🔥 لمستخدمي macOS وWindows وLinux
OpenScreen أداة مفتوحة المصدر لتسجيل الشاشة بجودة عالية، مع تكبير تلقائي للمؤشر، وإضافة أسهم ونصوص، وخلفيات جاهزة، وتصدير بصيغ MP4 وGIF وWebM، وكل ذلك بدون علامة مائية أو اشتراك.
Tonight I sat back from my screen and couldn't speak.
Computational engineering logic fused with modern CAD, and what came out wasn't just a design. It was a glimpse.
Every generation thinks it's living at the hinge of history. Ours actually is! The machines we imagined in dark theaters as kids are being born right now, not in billion-dollar labs, but at desks like mine. And you can start building them.
Gemma 4 now works on-device using React Native!
You can now run Gemma 4 fully offline in your cross-platform apps with local hardware acceleration:
⚡ Vulkan delegate on Android
⚡ MLX delegate on Apple Silicon
See Gemma 4's vision and tool-use capabilities in action, instantly reading a flyer and scheduling a calendar event, 100% on-device.
WEB SCRAPING JUST GOT A SERIOUS UPGRADE.
PixelRAG doesn't read HTML. It reads the page exactly like you do.
100% open-source.
Instead of parsing websites into plain text, it captures screenshots and lets a vision model retrieve answers directly from the pixels.
Why that's a big deal:
• HTML parsers silently lose information.
• Tables, charts, formulas, and layouts often disappear.
• Even changing the parser can swing RAG accuracy by ~10%.
PixelRAG skips that entire bottleneck.
It indexes what users actually see.
The team built a visual index of 30M+ Wikipedia screenshots, and it outperformed the strongest text-based RAG baseline by 18.1% on text-only QA.
Even cooler:
It includes a Claude Code plugin that gives Claude visual browsing.
Instead of scraping the DOM, Claude can screenshot any webpage, PDF, arXiv paper, or even your local app—and answer based on the rendered page.
The pipeline is surprisingly clean:
→ Render pages into image tiles
→ Embed with Qwen3-VL-Embedding (LoRA-tuned on screenshots)
→ Store in a FAISS index
→ Search visually
The best part?
Upgrade to a better vision model later, and you don't need to rebuild the index.
Because the index stores pixels, not parsed text.
Fully open-source under Apache 2.0.
GitHub: https://t.co/B7whbNg60s
I FOUND THE AI WORKFLOW EVERYONE WILL COPY IN 6 MONTHS.
Andrej Karpathy shared a simple idea: stop using AI just to write code use it as a second brain.
Point Claude Code at a folder. Drop in anything (articles, transcripts, PDFs). It reads, links, and files everything into a living wiki. The more you add, the smarter it gets.
Setup:
→ Install Claude Desktop + Obsidian (free to start; Claude needs paid for Code)
→ Create a new Obsidian vault = your brain
→ Enable Local REST API plugin + copy the key
→ Connect Claude Code to the vault via MCP (one command)
→ Test: ask Claude to list vault files
→ Have Claude interview you + save to /CLAUDE.md (permanent context)
→ Make project folders + per-project CLAUDE.md
→ Turn repeats into “skills” Claude can run
→ Connect tools (e.g., Google Calendar) via MCP
→ Schedule a daily run to organize new notes overnight
5 minutes to set up. After that, no more blank chats.
Bookmark this. Follow @cyrilXBT
Las escuelas deben ser lugares seguros. Siempre.
El ataque contra una escuela en Minab, Irán, ha segado la vida de niñas y adolescentes que estaban en clase.
UNICEF se suma al llamado del cese inmediato de la violencia y al pleno respeto del Derecho Internacional Humanitario.
Las escuelas #NoSonUnObjetivo. Nunca.
🎨: @unicefenespanol
❗🇮🇱 Israeli spyware company Paragon posted their surveillance software's control panel by mistake.
This is a major opsec flaw.
Paragon's spyware has been used to target journalists, government critics, and human rights defenders.