How many times do we have to give up our data before we learn that policy alone isn’t privacy?
If you want to protect your data, keep it local.
A quick explainer for everyone else.
>half the internet runs on PHP
>a big security bug dropped today
>basically hackers can take over websites just by sending one special request (crazy!)
>no password needed
>patches are out
>lots of sites are going down for emergency updates
>if you have a website, tell your host to update PHP right now
>if you don’t, just enjoy the random maintenance pages for now
AI just found two critical holes in PHP. The language behind most of the web.
If AI can find these, attackers can too.
Your data is only as safe as someone else’s server. Your AI doesn’t have to be.
Local is the way.
Effective Immediately, all organizations using PHP for their systems are advised to be prepared for an update addressing two critical vulnerabilities. PHP is widely used for web applications. These vulnerabilities often provide a direct path to the server's file system or database. If an immediate update is not feasible, it is recommend that organizations take the system offline until it can be fully mitigated. NSA will follow up shortly with more specific guidance.
Building a harness for many models means designing for two users.
The human. And the model.
Small models don't have the attention span big ones do.
Fewer tools, fewer mistakes.
tpae retweeted
6 of 7 local-first ideals, adapted for agents. The seventh is being built.
The full essay: xitter.cf/tpae/status/2112903853…
@martinkl @_adamwiggins_ @inkandswitch Peter van Hardenberg Mark McGranaghan
I read your essay in 2019 and never got over it. This one is because of you.
Do you own your software or rent it?
Simple test: if the company disappears tomorrow,
does it still work?
The more personal software gets, the less sense renting makes.
AI is about to be the most personal software we've ever used.
Every so often, go back to why you started.
Remind yourself.
Change without conviction means nothing.
tpae retweeted
@OsaurusAI does the muscle, Ornith does the hustle.
🙌 Two bros took all daily grind.
Check out the amazing work from Osaurus team!!
🔗 osaurus.ai
tpae retweeted
Congrats to the Ornith team. Ornith-1.5-122B is a monster.
It’s in Osaurus on day one. Running on a Mac, no cloud, no subscription.
One prompt. It built a flying car game.
Aloha!🌺 Meet Ornith-1.5-122B: No Tradeoff Between Strong and Fast
Stronger. Faster. Leaner. Ornith-1.5-122B delivers frontier-level capabilities with exceptional inference efficiency. With superior instruction following, stronger task decomposition, and deeper understanding of real-world workflows, it achieves state-of-the-art performance among open-source models of comparable size, while running on a single NVIDIA DGX card!!
Open weights under the MIT license.
Coding
✅SWE-bench Verified: 0.845
✅TB2.1: 0.813
✅CyberGym: 0.800
✅NL2Repo: 0.512
Agentic
✅HLE: 0.446
✅BrowseComp: 0.785
✅MCP-Atlas: 0.746
✅Toolathlon: 0.708
🤗Hugging Face: huggingface.co/ornith-ai/Orn…
We can’t wait to see what you build with Ornith-1.5-122B.
They said a single DGX card.
This is a MacBook. Ornith-1.5-122B, day one in Osaurus, no cloud.
Watch it build a flying car game from one prompt.
122B in 42.6 GB. 262k context. Vision and video. JANGH2 quant by @dealignai, shipped the same day their weights dropped.
github.com/osaurus-ai/osauru…
tpae retweeted
@NVIDIARTXSpark we’re building for @OsaurusAI for Windows (currently in beta) and have been making our own models tuned for specific tasks and our harness. (huggingface.co/OsaurusAI)
How can we work together to make sure RTX Spark users have the best local AI experience possible?
Anyway, this is the whole reason we build Osaurus.
If your files are worth $5M to somebody, maybe stop handing them out.
Model runs on your Mac. Wi-Fi off, still works. Nothing leaves.
Free, MIT, open source.
github.com/osaurus-ai/osauru…
