We just shipped NVIDIA-Verified Agent Skills 🔐
Skills make your agent more capable, but can also introduce vulnerabilities. Verified skills give you transparency into what a skill does, where it came from, what risks it carries, and whether it's been modified.
Every verified skill carries a skill card and is built on the agentskills.io open specification to work reliably across @claudeai Code, @openai Codex, and @cursor_ai.
Just remember that if you have to lie or bend "a little truth", it almost never stops there.
These little foxes compound, and eventually define the person you become.
Time will either collect its debt or pay you interest. Everything compounds, there's no such thing as stasis.
Tèmi retweeted
One of the most important papers in AI: a tiny brain-inspired 27M param model trained on 1000 samples outperforms o3-mini-high on reasoning tasks!
Still can't believe this tiny lab of Tsinghua grads gets 40% on ARC-AGI, solves hard sudoku and mazes.
We're still so early.
Me, looking at this, knowing I just began rewriting an entire app from React Native to Swift UI.
Every time I try to seriously use SwiftUI for a complex new feature, I come away slightly disgusted with myself for having wasted so much time implementing the most basic things that there's millions of libraries for in React.
Like, we're halfway to 2026 and you can still not create a smooth scrolling collection grid of items with a custom layout in SwiftUI. Sure, you can prototype an Apple-styled List™ or LazyVGrid™ view rapidly, but once you slap 500+ items in there and some real-world scenarios or styling, your app grinds down to a 10fps experience and you need a SwiftUI PhD and maybe another few days to claw back some 5-10 fps doing what feels like extremely hacky workarounds.
If you told that to someone in 2019, they would surely conclude that SwiftUI has failed as a framework.
Tèmi retweeted
Tokenization has been the final barrier to truly end-to-end language models.
We developed the H-Net: a hierarchical network that replaces tokenization with a dynamic chunking process directly inside the model, automatically discovering and operating over meaningful units of data










