Today we are officially launching The Local Intelligence Project. If you are interested in building your own AI sustainably, creating community, building resiliency against centralized power in tech, or dream of being involved in a digitally terraformed future, this is for you!
Best models for your hardware
- 4gb to 12gb vram -
VibeThinker-3B - smokes everything remotely close to its weight class. Challenging 30b models! Last version was also topping math benchmarks
https://t.co/RTchJFFTnV
- 12gb to 24gb vram -
Gemma-12B-coder
Built on top of an already strong model, reduced refusals and 262k context window trained on fable traces https://t.co/DVAhlQ7Y4n
- 24gb to 64gb vram -
Gemma-4-26b-diffusion
This model was already by far one of the most functional and capable models, now it’s hitting 500+ tok/s on consumer hardware! Smart AF made by Google deepmind https://t.co/mSaWPFpgXQ
Cohere North-Mini-Code 30B
A new coding model made by an already impressive lab, its priming worth a shot if you’re looking to test the limits of local coding https://t.co/gDPEj6lPAW
———
For those with 4x 6000s or 3x DGX Spark I think my GLM-5.2-REAP is worth a shot.
Lmk how it goes!
HICIERON UN CAPCUT GRATIS Y SIN MARCAS DE AGUA, Y TIENE 55K STARS EN GITHUB
CapCut te mete marca de agua, te bloquea funciones y encima te cobra suscripción. Un grupo de devs se cansó y construyó la alternativa open source.
→ Editor de vídeo completo, sin marcas de agua ni paywalls
→ Compatible con web, escritorio y móvil → Open source con licencia MIT
→ Servidor MCP incluido para agentes de IA
→ Se está reescribiendo en Rust desde cero con API, plugins y scripting
Se llama OpenCut y es exactamente lo que CapCut debería haber sido desde el principio.
Te lo explico abajo (link de la repoo también) ⬇️
When you grew up in the 80’s and god gives you @threejs and ai in your mid 40’s.
As a creative, I now have a creative outlet with these technologies that I thought I’d never have. This wasn’t a ‘one shot’ game by any means, took some work…. But what has come out has amazed my kids and I. 8 player coop solar system gravity racing with a banging soundtrack.
People slate ai for coding a lot, but for me it has enabled me to put ideas into reality far quicker than I could have imagined.
From a CNC parametric modelling app to a game, it’s astonishing what can be made.
I cannot wait to get this out there for others to enjoy. 🙏 to those at @threejs for the effort, without which this wouldn’t exist.
Insane, Yeah he really did it, he made a RAM at home in his backyard shed.
While big tech cries about RAM shortages Man builds functional DRAM from scratch using homemade sputtering and lithography tools.
20-bit memory cell array, 12pF capacitance.
Turned it into a legit Class 100 cleanroom and fabricated memory cells himself. 5x4 memory cell array fabricated,This is the first RAM ever made at home.
Drug lab vibes, semiconductor god mode.
So apparently Google ships a Gemini Nano 4B LLM (context limit : 9216 tokens) baked into chrome
I tried to expose it out as an OpenAI-compatible API for my local, so basically no API key, no external network calls, no need of ollama
Demo and repo below ↓
🤯 This is a website, a simple web-based game built with WebGL and Three.js.
Website: https://t.co/G0cDgaWzKe
It's honestly surprising how far web development has come.
If you’re still sending raw JSON into your LLMs, you’re burning tokens, latency, and budget!
Try TOON (Token-Oriented Object Notation).
Clear like YAML, compact like CSV:
• 30–60% fewer tokens
• Up to 50% lower costs
• Shines for tabular data.
Free and Open source 🧵↓
56,000+ tokens/sec at just 80 MHz. 🤯
I burned a full Transformer with KV cache into a custom chip. Designed gate by gate as a 100% digital integrated circuit. Prototyped on a FPGA. (No GPU. No CPU)
Just pure digital silicon running @karpathy microGPT, spelling out names on a tiny LCD.
This is GateGPT 👇
This is really big news. Google introduced the Open Knowledge Format (OKF) - a standardized way to store information in a directory of markdown files. Makes it really easy to make a digital brain that agents can use.
These files can serve as a living wiki. You can give agents the ability to query them or edit them. They can interlink.
Seems to me this could replace Notion or Obsidian. I can think of so many uses for this.
Google's blog post: https://t.co/DqSjg4UpvH
An easier to understand explanation is the SPEC.md file:
https://t.co/A3qSz3Tfas
I gave those two links to Antigravity and asked how we could use it for any of the projects we're working on. It came up with so many ideas. I would imagine Claude Fable 5 would whip up some pretty amazing things based on this system.
Currently creating an OKF library of our pepper garden. It's going to be a fun weekend.
“don’t train your own model” is common ai advice. it's wrong. your token bill's the proof.
today, we’re excited to launch castform into open preview. castform is the easiest way for you to train your own model, on your own data.
open-weights models are performant and much cheaper. when trained on your task & proprietary data, they beat closed models. the thing standing between you and that was weeks of plumbing & years of ml expertise.
with castform, model training is as simple as prompt engineering. @castformai
bring your agent traces or raw corpora. castform turns it into training data, picks the right algorithmic recipes, manages gpus, and gives you an ide to watch and chat with your model as it learns.
see what you can build with castform👇
🚨 WTF?! A USS Liberty survivor drops a massive bombshell.
He confirms they were threatened with prison or worse if they ever spoke about the Israeli attack.
Another veteran explicitly states "Israel owns us" after being ignored by Congress for 59 years. Total betrayal!
A GUY AT GOOGLE DEEPMIND MADE AN ISOMETRIC PIXEL-ART MAP OF NEW YORK CITY AND PUT IT ON THE OPEN WEB FOR FREE
it's called https://t.co/8fAASvEmXT
you open the tab and the city is just sitting there in classic SimCity 2000 isometric pixel art. you scroll. and it keeps going. and going.
i zoomed in on midtown and i could read the H&M signage in times square. in red. as actual pixel-art letters on the side of a building.
i could see the crystalline spire of the Bank of America Tower poking out of a clump of skyscrapers. individual rooftop HVAC units. tiny green roof gardens. the little driveway loops in front of the hotels.
he estimates the map needs roughly 40,000 tiles. nothing is a placeholder.
the guy who made it is Andy Coenen, a senior staff engineer at Google DeepMind. he is not a pixel artist. by his own admission he is "a former electronic musician."
what he actually did is kind of insane:
> pulled NYC's geometry from the Google Maps 3D tiles API
> fine-tuned an open-source image model (Qwen-Image-Edit) on ~40 hand-paired examples of "satellite tile → pixel art tile"
> spun up 50 parallel instances on rented GPUs and generated tens of thousands of tiles in a few hours
> the fine-tune cost him 12 bucks
his own stated mission for the project, verbatim, is one sentence: "what's possible now that was impossible before?"
apparently the answer is "one engineer can pixel-art most of a metropolis for the price of a sandwich."
and the wildest part to me is he didn't sell it. no signup. no paywall. no NFT. you open the URL and the city is yours to wander.
the post landed at 1,325 points on Hacker News and topped bestofshowhn's 2026 list.
we live in a timeline where a senior engineer at one of the largest AI labs on earth spent his nights pixel-arting Manhattan for fun and then gave it away.
the internet is healing.
The man who killed the $10,000 GPU myth.
He did it alone, from Bulgaria, with one C file. 🤯
>Meet Georgi Gerganov.
>Bulgarian developer. Nobody had heard of him.
>In March 2023, Meta’s LLaMA model leaked online
>Within days he wrote a single C file
>Called it llama.cpp
>It ran a full AI model on a MacBook. No GPU. No cloud.
>The entire AI industry said you needed $10,000 GPUs to run LLMs 🔥
>He proved you didn’t. On a laptop. Alone.
>Also built whisper.cpp ~ same thing for voice AI
> His code is the foundation of Ollama, LM Studio, and GPT4All
>107,000+ GitHub stars. Fastest open-source Ai project to hit 100K ever. 🚀
>In 2026 Hugging Face hired his entire team
>Still ships code. Still open source. Still free.
Whenever you run AI locally, you’re running his work.
Absolute Legend 🐐
The scariest AI future isn't Terminator.
It's five companies quietly deciding what the model knows, who it serves, and what it refuses to say.
The antidote isn't regulation.
It's infrastructure that no one can own
Join us: https://t.co/PxyGCgtFeG