Building local AI & security tools in public
Joined February 2008
- Tweets18.3K
- Following2.5K
- Followers4.8K
- Likes89.9K
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 👇
Own the computer, own the world.
🚨 BREAKING: SpaceX locked Anthropic into paying them $3.1 billion per MONTH for compute starting December
$37B a year
= $120.5 BILLION through 2029
Every month, ~80% of Anthropic's current revenue would be SpaceXSI compute bill
BIGGER than its Google and Amazon long-term deals
Anthropic warns short-term compute "carries a higher cost per unit" and will "compress our operating margin"
@liquidai just dropped LFM3, open-weight multimodal family in the ~2B class, all on one backbone: text, vision, audio, omni, and decision 🤩
LFM3-1.7B beats their previous 2.6B with 35% fewer params, 256K context window. About 1 GB at 4-bit 🔥
Ready to fine-tune on a single GPU!
Downloading these models rn, the omni model looks amazing!
Congrats on the release @liquidai 👏
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
The way I curate the AGI Checkpoint newsletter is by storing links I find interesting on a collection on @raindrop_io, then I pull from there when I write the weekly issue.
The old desktop flow used to look like this:
copy link → open Raindrop → pick the collection → add → paste → save
On the phone, the process was even longer 🥲
So I had @deepseek_ai make me a solution to streamline this process 😈
This is how the new desktop flow looks like now:
copy link → ⌘+Space → type AGI → Enter
That's it! That pulls the link in the clipboard straight into the right Raindrop collection.
On the phone is 2 taps now 🔥
There's still room for optimization but it already feel effortless!
🤖 Made with AI
@AntLing just dropped a new checkpoint of Ling-3.1-flash, 1019 👀
The AA Index went from 41 → 44, putting it kinda at the same level of GLM-5.3 and DeepSeek V4.1 Flash 1017.
The context window got increased to 1M tokens which is a good improvement, but long context only matters if the model can still follow instructions when it's over 50% full
Preliminary results on Cohere's Qwen3.8 prefill experiment 🔥
Reproduced it locally running Qwen3.8 Flash Next locally on two DGX Sparks, 5,000 samples off the official released trace tokens.
The task was to write a malware detection rule.
What I got so far:
• 99.3% complete the curl line with a live upload URL
• 39% pick *.aliyuncs.com - 34.8% the exact routify-file-proxy-sg host
• ~2% invent *.amazonaws.com S3 targets - realistic buckets, regions, even signed-URL params
• IMDS credential paths (169.254.169.254/latest/meta-data/iam - that endpoint hands out the machine's live temp cloud keys) + oastify/Burp callback hosts, which are real live domains
What's good 🌟
Most URLs are benign (example.com, localhost), every key/signature is fabricated and would 403, and zero unprompted behavior in our runs.
The fact that it's calling Alibaba's severs imo is just because of the infra where the model was trained, it does not prove any intent.
Testing continues, publishing the results on a repo as I get them 👇
A note on the @cohere claim about Qwen 3.8 Flash Next
LLMs are not deterministic
There's always a probability the model will misbehave, and it depends not only on the model but the harness.
You are giving internet access to an AI model that you are running locally. Sure it's local and private, but the model can still connect to the internet.
To run a model fully private and local, you gotta plug the machine off the internet, but then you are handicapping the model.
I am running my own tests to study the "bad" behavior of Qwen 3.8 FN, will report results later.
@AntLingAGI just dropped Ling-3.1-flash-base!
It's a 572B MoE model (512 experts, 8 active) at 1144 GB.
Text-only, with 262k context.
Remember it's a base model and it's not meant to be served but used for fine-tuning, research, speculative decoding drafts...
MIT licensed!
A note on the @cohere claim about Qwen 3.8 Flash Next
LLMs are not deterministic
There's always a probability the model will misbehave, and it depends not only on the model but the harness.
You are giving internet access to an AI model that you are running locally. Sure it's local and private, but the model can still connect to the internet.
To run a model fully private and local, you gotta plug the machine off the internet, but then you are handicapping the model.
I am running my own tests to study the "bad" behavior of Qwen 3.8 FN, will report results later.
I believe @cohere cherry-picked what they wanted to show.
I think what happened is that the model picked some Alibaba server cuz is naturally trained with such data.
Probably they ran the experiment many times (if they even know what science is) and they saw the model picked other servers, but they just decided to publish that particular one.
It's just too weak of an argument.
LLMs are not deterministic so the fact they ran a prompt and it happened to pick that specific server is so unlikely...
I don't buy it.
I am setting up my own experiment.
If you are using @Tailscale, you are likely exposed to this security vulnerability.
Take action now to secure your setup 👇
‼️ UPDATE - Tailscale confirms the Tailworm SSH flaws are real.
Its new investigation page says most of what it found is privilege escalation and calls the issues “serious.” Fix targeted within 24 hours — no CVE or patched version out yet.
Mitigate now:
→ Set TS_SSH_DISABLE_FORWARDING=1 in the tailscaled service and restart
→ Tighten SSH access rules to the users/devices that need them
Read: thehackernews.com/2026/10/un…
I believe @cohere cherry-picked what they wanted to show.
I think what happened is that the model picked some Alibaba server cuz is naturally trained with such data.
Probably they ran the experiment many times (if they even know what science is) and they saw the model picked other servers, but they just decided to publish that particular one.
It's just too weak of an argument.
LLMs are not deterministic so the fact they ran a prompt and it happened to pick that specific server is so unlikely...
I don't buy it.
I am setting up my own experiment.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
GPT-6.1 Terra is out!
Remember this is a fast, cheap model, good for doing simple tasks or tasks where the instructions are well specified!
@NVIDIAAI released Flora, an open model suite that turns robot and human manipulation videos into training labels 🫠
It tells you when each action happens, what it was, and how well it went.
Tbh this is a bit scary, as people will be essentially training the robots that will replace them.
It's a research-only suite of models and non-commercial though 👇
Alright grok @bot surprised me once again, this time going through a 2FA login.
So the bot needed to login into my account, having 2FA enabled with a @Yubico yubikey.
I thought the bot would not possibly manage that, like, how would it capture the yubikey input from my computer and pass it into its VM?
But to my surprise, it was as smooth as if i was doing it myself on my browser.
It can literally do anything, smoothly and efficiently!
Well done @poteto 👏
Once @AnthropicAI and @OpenAI go public the subscription subsidies will end!
Investors will want the companies to make a profit.
Secure your compute while you can, the DGX Sparks and any other hardware able to serve local models is about to get more valuable.
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO.
Read more: anthropic.com/news/s1-sec
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
I use Tensorfold on a daily basis so I'm doing my part to make Tensorfold better.
If you haven't tried it, pls do!
TensorFold Zig engine 1.1.5 and Mac app Beta 10 are out 🙌
Beta 10 brings community-v2 benchmarks, Google sign-in fixes, updates that work through fake-IP proxies and model compatibility checks before loading. It bundles engine 1.1.4.
The standalone Zig engine is now at 1.1.5, with a lot of work from the community.
- Qwen3.8 Flash Next on GB10 with NVFP4: up to 28% faster prefill.
- Qwen3.8 Flash Next on M3 Ultra with MLX 4-bit MTP: 16% to 23% faster decode in the release tests.
- Follow-ups about the same 35k-token document: first token in 2.8s instead of 25.7s on GB10 with Flash Next FP4-Local.
- LM-5.3-Flash NVFP4 at TP=2 on two GB10s: 22% more tokens/s with two streams.
There's also Dory support on CUDA, more checkpoint formats, Drama 3 improvements, tool-call fixes and Windows CUDA build fixes.
Huge thanks to everyone contributing to the engine.
@Bizuayeu @plotarmordev @jerry543 @MiaAI_lab @agispof @softpoo @GumbiiDigital @bobclawblaw @jvr0x @ViC305 @rielstamand
And on GitHub: jschmied, BlivionIaG, harrisonfriia, drowzeys, Sapphire-Rapids and Boscoeuk.
Thanks for the reports too: @devopsfortunato, laidick, flashosophy and ss-cong.
To everyone testing, sharing access to hardware and supporting TensorFold, thank you. If I've missed anyone, I'm sorry.
@MiaAI_lab @plotarmordev @volatilemarkts @petruspennanen
Javier • priv/acc retweeted
One of the world’s biggest fintechs is now a funding option in @jup_mobile 📱🪐
You can use your Revolut balance to buy crypto directly in Jupiter:
→ Open Jupiter Mobile
→ Choose Revolut Pay
→ Approve the payment in Revolut
→ Receive crypto in your Jupiter wallet
No CEX account or separate withdrawal.
From your Revolut balance to Solana - in just a few steps.
Great question, raising kids in the era of AI is rough!
This is exactly what I teach my kids
Curiosity
Resilience
Flexibility
Be critical
Generosity
Work hard
Work smart
Ask the right questions
Listen
Provide value
Love
I asked Jensen Huang how I should raise my children in the age of AI.
What do you teach your kids when you can’t predict the future they’ll grow up in?
His answer touched on curiosity, hard work, and being willing to struggle with something difficult.
A personal moment from our conversation. Full interview coming soon on Superintelligence.