Building local AI & security tools in public
Joined February 2008
- Tweets18.5K
- Following2.5K
- Followers4.9K
- Likes90.1K
Pinned Tweet
Javi0.3 is out 🚀
Two tiny agent models, 4B and 2B, that think from first principles and use its tools effectively to find the answer to your prompt. They fit under 4 and 2GB!
Everything about this model, every mistake, with tables and training data 👇
𝗔𝗚𝗜 𝗖𝗵𝗲𝗰𝗸𝗽𝗼𝗶𝗻𝘁 #𝟰 𝗶𝘀 𝗼𝘂𝘁 🚀🚀
A few highlights of this week are:
𝗕𝗶𝗴 𝗹𝗮𝗯𝘀 🏢
→ @GoogleDeepMind's Gemini 4 Carbon takes #1 spot on the Artificial Analysis index
→ @MistralAI released new Le Chonk weights
𝗖𝗼𝗺𝗺𝘂𝗻𝗶𝘁𝘆 👥
→ @ViC305 squeezed @Meta's 816B Muse into ~120 GB at 1.13 bpw, with 96.48% top-1 token agreement
→ @divagr1925 built a RISC-V CPU inside Qwen3.5-4B's weights, using its Gated DeltaNet layers as RAM. It runs C and MicroPython in its forward passes
→ @plotarmordev cut Fish Audio's Drama 3 from 36.95 to 19.35 GiB in NVFP4, with no measurable quality loss on one Spark
More builders who shipped cool stuff this week and you should follow 👇
@jayleaton: LocalRouter, a full local AI stack on one DGX Spark
@u1tra_instinct: Ghost Machine, near-instant speech-to-speech agents on TensorFold
@VectorCrossProd: speech-to-text running on an ESP32-P4 chip
@roy_eisenstadt: DCOCO, speculative decoding up to 6.6x faster
@ashxhart: more improvements to TensorFold
@MiaAI_lab: no intro needed, the legend behind many spark recipes
@ch0mpaa: TensorFold on Snapdragon
Exciting times 🤩
This is incredible.
Neuralink is not only helping people who lost the ability to speak but also enhancing the way humans communicate with computers.
The way we interact with our devices is way too slow. Even if you type fast, it's still slow.
Nowadays I find myself speaking more and more often to the computer via a speech-to-text tool I built and that helps but imagine how it would be when you can just think of something and that idea is directly transferred to the computer.
This will have s huge impact on how we shape our AI models, as thoughts are very chaotic. Teaching models interpret that mess is gonna be interesting.
Can it be that in the future, speech becomes optional? Scary and fascinating, as this is something that makes us humans.
Such a terrible idea to have sovereignty over your intelligence.
Of course it's terrible for his subscription business 😅
Own your compute!
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
4th DGX Spark secured ✨
Going to write an article on how to set them up without a switch soon!
🤖 Made with AI
The weights for @MistralAI Le Chonk are finally out!
I am liking what I see on their blogpost and the scores look really interesting, hopefully not benchmaxxed!
And they gave us an NVFP4 quant! 🎉
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity.
💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis.
🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others.
Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it.
Read more from our team and customers 👇:
mistral.ai/news/mistral-larg…
Get the weights 👇
huggingface.co/mistralai/Mis…
NVFP4 weights 🎉
huggingface.co/mistralai/Mis…
Qwen4-Flash and Qwen4-27B, I like what I see @Alibaba_Qwen 🤩
I am ready
Just applied to the @SpaceXSI ambassador program 💪
Ambassador applications for @spacexsi / @cursor_ai are open
Details + apply here: anysphere.typeform.com/space…
Javier • priv/acc retweeted
YC builders building on Solana receive $50k + support from @SolanaFndn
Apply to YC by November 2: ycombinator.com/apply
At @ycombinator, we're excited to fund more crypto startups solving financial needs including AI capital raising, stablecoins, trading, and agents.
Here's our latest crypto request for startups. Our next application deadline is Monday.
I started to become more bullish on @GoogleAI after seeing what they are doing with their recent models 🔥
@AnthropicAI has done incredible with their models but you know what they lack?
𝗖𝗼𝗺𝗽𝘂𝘁𝗲
Google owns it. @SpaceXSI owns it.
Anthropic and OpenAI rent it.
The future belongs to those owning their compute, like us running Local AI 😉
🤖 Made with AI
Javier • priv/acc retweeted
CUDA Live: Introduction to CUDA Rust nitter.cf/i/broadcasts/1RKZzBapW…
A year from now we will interact with computers very differently.
AI is able to control the computer, files, browser, and can process data way faster than humans.
You gotta be good at the high level picture to direct the agents to solve the right problem.
Exciting times 🤩
Your local LLM can now take control of a browser and complete tasks autonomously via Unsloth!
Qwen3.8-27B successfully controlled the browser to complete tasks like chess, painting, with no human intervention.
GitHub: github.com/unslothai/unsloth
Guide: unsloth.ai/docs/basics/brows…
€25k for the Mac Studio in Europe. 512GB M5U, 1 TB storage!
If you wanna bump that to 4TB+ it lands at €28~30k 🤯
Nemotron 3.5 Transcribe looks very capable!
Will try it out next time I'm adding subtitles to my videos 😍
Good one @NVIDIAAI 👏
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️
An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
Watch it keep track of who said what when everyone’s talking at once.
1/ 🧵
With AI, the possibilities are endless.
Think outside the box 🤩
Full Mario 64 in Mixed Reality! Running in real 3D (no stereoscopic crap) natively at 90FPS and also running on my desk! 😋
Again, Opus 5.5 + PICO CLI combo made it quite straightforward.
Such a bliss to play an oldie at 90 FPS in 3D.
Disclaimer: I own an original physical copy of Mario 64. This project is not associated with Nintendo or PICO.