George Chiramattel
@georgeck.me
Lifelong learner | Exploring topics mostly related to technology & sharing knowledge. Previously: Intuit Inc.
My talk at @localfirstconf.com is up, in which I argue that local-first has a role to play in improving the resilience of democratic societies in an era of geopolitical tensions. I'm pretty happy with how this talk turned out www.youtube.com/watch?v=--Ru...
Martin Kleppmann — Local-first in an unstable world
YouTube video by Local-First Conf
youtube.com
This is the second installment in our blog series on Agentic systems.
Latent Signal@latentsignal.org · 3w agoBuilding an agentic system? Start with a direct model call. Add an SDK, gateway, runtime, or channel abstraction only when it solves a recurring engineering problem. Part 2 of our series on agentic systems: latentsignal.org/blog/agentic...
Inkling: Our Open-Weights Model View Article | Discussion + AI Summary Summary of HN discussion 🧵👇
Inkling: Our open-weights model
Our first open-weights model: multimodal, Mixture-of-Experts, with controllable reasoning effort. Available to fine-tune on Tinker.
thinkingmachines.ai
Across different agent projects, we kept running into the same questions - which pattern fits the problem, what options are available, and what trade-offs come with each one? @georgeck.me and I are putting what we learned into a new series, starting with the Autonomy Spectrum of agentic systems.
Latent Signal@latentsignal.org · 3w ago“Agentic systems” can refer to very different kinds of systems. In our new series on building agentic systems, the first post maps the autonomy spectrum—from simple model calls to systems that plan, use tools, and act with limited supervision. 🧵 1/3
1-bit LLM Bonsai 27B, shrinked from 54GB to just 3.8GB (-93%), runs locally in your browser Models: huggingface.co/collections/... Demo: huggingface.co/spaces/webml...
Bonsai 27B - a prism-ml Collection
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
“Agentic systems” can refer to very different kinds of systems. In our new series on building agentic systems, the first post maps the autonomy spectrum—from simple model calls to systems that plan, use tools, and act with limited supervision. 🧵 1/3
We’re releasing Gemma 4 NVFP4 quants that run 1.5× faster on your GPU. Gemma-4-12B NVFP4 works on 11GB VRAM. 26B-A4B hits 13K tok/s (B200). Unsloth NVFP4 enables faster, more accurate 4-bit Blackwell inference. Blog: unsloth.ai/docs/basics/... Gemma NVFP4: huggingface.co/collections/...
Local, CPU-Friendly, High-Quality TTS (Text-to-Speech) with Kokoro View Article | Discussion + AI Summary Summary of HN discussion 🧵👇
Local, CPU-Friendly, High-Quality TTS (Text-to-Speech) with Kokoro
Just a few years ago, realistic local speech generation seemed unimaginable. Today, its quality is exceptional and, crucially, it delivers these results without compromising privacy.
ariya.io
Bluesky isn't just growing. It's changing how reach works! We pulled 6 facts straight from @bsky.app's own blog and transparency reports to help you decide if it's worth your time in 2026. 🧵 ↓
Cat and Bluesky Butterfly Animation
ALT: Cat and Bluesky Butterfly Animation
static.klipy.com
Big news for DwarfStar users: I got DeepSeek v4 Flash and GLM 5.2 working with Tensor Parallelism across 2 M5Max 128GB MacBooks via RDMA. It is especially interesting for GLM since otherwise, fully resident, can't fit a machine that money today can buy... Now it can.
🛠️ hunk 0.17.0 is out Patches from 10 new contributors, w/ small fixes & quality-of-life improvements like: - official homebrew install (brew install hunk) - launch hunk w/o menu bar - better session reliability - esc no longer quits, proper transparent bg colours, & more
How to sequence your own DNA at home View Article | Discussion + AI Summary Summary of HN discussion 🧵👇
How to sequence your own DNA at home
I have now sequenced my own genome 5 times with an Oxford Nanopore Technologies MinION. This means collecting them from a swab, prepping them for sequencing, running them through a sequencer, then doing analysis over them.
bradleywoolf.com
Synthesis is harder than analysis View Article | Discussion + AI Summary Summary of HN discussion 🧵👇
Synthesis is harder than analysis
Over the years, mathematicians, logicians and computer scientists have developed various calculi. If you have a background in computer science, you’ve likely heard of the lambda calculus, a m…
surfingcomplexity.blog
Fable created novel 4D splat format View Article | Discussion + AI Summary Summary of HN discussion 🧵👇
.splat4d
A streamable 4D gaussian splat format with tunable error bounds. 16–58× smaller than raw; streams and seeks over plain HTTP Range requests.
adamraudonis.github.io
Explanation of everything you can see in htop/top on Linux (2019) View Article | Discussion + AI Summary Summary of HN discussion 🧵👇
htop explained
Explanation of everything you can see in htop/top on Linux
peteris.rocks
Valve open-source the Steam Machine e-ink screen so you can make your own View Article | Discussion + AI Summary Summary of HN discussion 🧵👇
Valve open source the Steam Machine e-ink screen so you can make your own
While Valve will not be making and providing their own e-ink display for the Steam Machine, they have opened it up so anyone can now do it.
gamingonlinux.com
The Hitchhiker's Guide to Agentic AI: From Foundations to Systems (603 pages) arxiv.org/abs/2606.249...
I ported Kubernetes to the browser View Article | Join the HN Conversation Summary of HN discussion 🧵👇
I ported Kubernetes to the browser | ngrok blog
Almost 100,000 lines of LLM-generated code in 2 months, and none of it is slop.
ngrok.com
The gap between open weights LLMs and closed source LLMs View Article | Join the HN Conversation Summary of HN discussion 🧵👇
Prediction: A Frontier Open Source LLM Will Be Released On 3rd December 2026 | Doubleword
Using artificial analysis benchmarks we try to predict when an open source LLM will be released that matches frontier LLMs.
blog.doubleword.ai
We also offer a half-day in-person session for small product, design, and engineering groups; even folks who don't code day-to-day leave with a working prototype and a shared playbook. 👉 Reach out at: latentsignal.org/workshop/pri...
Private team workshop — Latent Signal
Bring the Claude Code agentic coding workshop to your team in person. Tell us about your team and we will get back to you with a tailored proposal.
latentsignal.org
Latent Signal@latentsignal.org · 2mo agoWe've opened our free and self-paced Agentic Coding workshop. Build a real-time, multi-user Idea Board with Claude Code - deployed to Vercel on your own URL in 3-4 hrs. AI foundations + 9 hands-on steps. For anyone who wants to understand how agents actually behave. latentsignal.org/workshop
We've opened our free and self-paced Agentic Coding workshop. Build a real-time, multi-user Idea Board with Claude Code - deployed to Vercel on your own URL in 3-4 hrs. AI foundations + 9 hands-on steps. For anyone who wants to understand how agents actually behave. latentsignal.org/workshop
Workshop - Latent Signal
Free, self-paced workshop on agentic coding. Build and deploy a real-time Idea Board app from scratch in 3-4 hours.
latentsignal.org
How many of the 170k English words do you know? View Article | Join the HN Conversation Summary of HN discussion 🧵👇
VocabOwl - Scientific Word Count
vocabowl-870366514258.us-west1.run.app
To test GLM 5.2, we utilized it in opencode.ai with the following task: Migrate the Umami project from Postgres to SQLite. This task is not particularly challenging because it has been attempted before, and Umami utilizes @prisma.io, which already supports SQLite. github.com/umami-softwa...
GitHub - umami-software/umami: Umami is a modern, privacy-focused analytics platform. An open-source alternative to Google Analytics, Mixpanel and Amplitude.
Umami is a modern, privacy-focused analytics platform. An open-source alternative to Google Analytics, Mixpanel and Amplitude. - umami-software/umami
github.com
The rollout of GLM Coding Plan from Z.ai appears a bit rushed, to be honest. There’s a 5-hour quota, but the usage dashboard doesn’t indicate when the quota will expire. Or am I misunderstanding something?
GLM-5.2 can now be run locally! 🔥 The 2-bit model retains ~82% accuracy after we shrunk it from 1.51TB to 238GB (-84% size). Run on a 256GB Mac or RAM/VRAM setups. GLM-5.2 is the strongest open model to date. Guide: unsloth.ai/docs/models/... GGUF: huggingface.co/unsloth/GLM-...
If you haven’t tried @pushover.net, give it a shot. It’s incredibly easy to set up, and the best part is that it’s a one-time payment.
Latent Signal@latentsignal.org · 2mo agoI didn’t want to sit on a Show HN page waiting for new comments, so I asked Gemini to write a small script that watches Firebase and sends Pushover notifications to my phone. I ran it once and left it alone. This morning I woke up to a stack of comment alerts.
I am sure this will be useful for others, so I turned it into a small GitHub project: github.com/latentsignal...
GitHub - latentsignal-org/hn-comment-watcher: Watch a Hacker News story for new comments and send Pushover notifications when replies appear.
Watch a Hacker News story for new comments and send Pushover notifications when replies appear. - latentsignal-org/hn-comment-watcher
github.com
Left: Our Projects page in prod Right: Clone built by GLM 5.2 + pi with a screenshot of the page Result - Not the best output, but close. However, we're impressed that GLM built it without image support - thanks to pi asking gemini CLI to extract colors, font, text from the screenshot upfront.
