Isaac @superstructor

In Claude Code, we’ve recently launched HTTP hooks, easier to use and more secure than existing command hooks! You can build a web app (even on localhost) to view CC’s progress, manage its permissions, and more. Then, now that you have a server with your hooks processing logic, you can easily deploy new changes or manage state across your CCs with a DB. How do HTTP hooks work? CC posts the hook event to a URL of your choice and awaits a response. They work wherever hooks are supported, including plugins, custom agents, and enterprise managed settings. Docs: code.claude.com/docs/en/hooks

Introducing a new method to teach LLMs to reason like Bayesians. By training models to mimic optimal probabilistic inference, we improved their ability to update their predictions and generalize across new domains. Learn more: goo.gle/4ue4eqj

if (GPU == Blackwell && precision == NVFP4) { DeepSeek.enable("BEAST_MODE"); } @Microsoft partnered with us to run DeepSeek‑V3.2 on GB200 NVL72 using NVFP4 + TensorRT LLM — and the numbers are impressive: ⚡ Up to 2.5x lower latency vs. H200 at the same settings ⚡ Serve 16x more users on the same GPU footprint ⚡ Near‑FP8 accuracy thanks to NVFP4’s 4‑bit quantization Read the details on the Microsoft Foundry blog: nvda.ws/4b6I8x6

Jon Seager, VP of Engineering at Canonical, recently presented about Rust adoption at scale with Ubuntu at @RustNationUK. 📹 Give it a watch to learn about the projects driving Rust adoption at Canonical, reasons behind the initiatives, and what’s next. youtu.be/WOA4GA8tAWA?si…

New ByteDance paper shows how an AI learned to write CUDA hardware code so well it beats standard compilers at their own game. This system creates custom software components that run up to 100% faster than traditional automated tools. Writing instructions for AI hardware is incredibly difficult because human programmers must perfectly manage memory and complex math operations. The researchers solved this by giving an LLM a secure testing environment where it can write code, run performance checks, and learn from its mistakes through continuous trial and error. Since there is very little training data available for this specific programming language, the team forced the system to combine basic math operations into thousands of new practice problems. Instead of just rewarding the AI for getting the right answer, the system gives high scores only when the code runs significantly faster than existing baseline software solutions. When tested against top commercial models and built-in optimization tools, this custom agent consistently found unique shortcuts that made the hardware run faster. This proves that language models can move beyond just typing out basic code to actively engineering hardware-level performance upgrades. ---- Paper Link – arxiv. org/abs/2602.24286 Paper Title: "CUDA Agent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generation"

An NVIDIA powered farming machine uses Al vision and precision lasers to eliminate weeds in milliseconds without herbicides and without harming crops, a potential shift toward chemical free agriculture

“If anyone can refute me—show me I’m making a mistake or looking at things from the wrong perspective—I’ll gladly change. It’s the truth I’m after, and the truth never harmed anyone.” – Marcus Aurelius

Today's Google Cloud livestream focused on Agent Skills in Gemini CLI! 🎓 Agent Skills help you provide specialized expertise while being context efficient. Missed it live? Catch up below 👇

Google Cloud Tech @GoogleCloudTech

4 months ago

On this week's livestream, we’re exploring Agent Skills like the recently released Agent Skills for Firebase with the help of Gemini CLI.

4 20 130 34K 60

A small Qwen3.5 from-scratch reimplementation for edu purposes: github.com/rasbt/LLMs-fro… (probably the best "small" LLM today for on-device tinkering)

Today, we’re introducing Gemini 3.1 Flash Lite (in preview) ⚡️ Now available via the Gemini API, our fastest and most cost-efficient Gemini 3 series model: - Features dynamic thinking for scaled reasoning - Delivers enhanced performance at a lower cost (priced at $0.25/1M input tokens and $1.50/1M output tokens) - Is built for high-volume dev workloads at scale - Outperforms 2.5 Flash in quality and speed - And more Start building → goo.gle/4lpiwQV

Introducing Agents UI, an open-source @shadcn component library for building polished React frontends for your voice agents. Audio visualizers. Media controls. Session management tools. Chat transcripts. All wired to LiveKit Agents. Install via the shadcn CLI and own the code.

You can now fine-tune Qwen3.5 with our free notebook! 🔥 You just need 5GB VRAM to train Qwen3.5-2B LoRA locally! Unsloth trains Qwen3.5 1.5x faster with 50% less VRAM. GitHub: github.com/unslothai/unsl… Guide: unsloth.ai/docs/models/qw… Qwen3.5-4B Colab: colab.research.google.com/github/unsloth…

"If something is humanly possible, it's attainable by you too." - Marcus Aurelius

Prof. Donald Knuth opened his new paper with "Shock! Shock!" Claude Opus 4.6 had just solved an open problem he'd been working on for weeks — a graph decomposition conjecture from The Art of Computer Programming. He named the paper "Claude's Cycles." 31 explorations. ~1 hour. Knuth read the output, wrote the formal proof, and closed with: "It seems I'll have to revise my opinions about generative AI one of these days." The man who wrote the bible of computer science just said that. In a paper named after an AI. Paper: cs.stanford.edu/~knuth/papers/…

Gemini 3.1 Flash-Lite has landed. It’s our most cost-efficient Gemini 3 series model yet, built for intelligence at scale. Here’s what’s new 🧵

Voice mode is rolling out now in Claude Code. It’s live for ~5% of users today, and will be ramping through the coming weeks. You'll see a note on the welcome screen once you have access. /voice to toggle it on!

Tesla will be one of the companies to make AGI and probably the first to make it in humanoid/atom-shaping form

Whenever someone wrongs you, remember the punishment you want them to suffer is the life they already have. You can’t inflict more pain on someone than an immoral person inflicts on themselves.

In a post scarcity world, there will still be scarcity. People will still want scarce human stuff Ex: We watch humans play chess even though computers could beat them. Taylor swift will still have people who want to go to her concerts. It’ll be an “and” not an “or” situation

Remote Control is now available in Research Preview for Max users, and coming soon to Pro users. Run `claude rc` to get started. Docs: code.claude.com/docs/en/remote…