Renting AI by the token works if your budget is nearly unlimited.
OpenClaw's creator estimates $10M in tokens in one month, and "not all of that is useful."
https://t.co/Dku2YW65wp
Most of a company's AI work can run on open models it hosts itself, says IBM's chief executive.
He told investors they're "now 90 percent as effective" as top models, at a fraction of the price. His question: "how many tasks require that?"
https://t.co/Hn4qmEW93k
Your AI should know who checked what it reads, and that record should sit in files you own.
OKGF does it in plain Markdown: who approved a page, who it's for, how good it is, where it came from. Plus a signature you can check.
https://t.co/pRK7IaPYnW
You can't own code you can't trace.
So in ASDD, the open process we build Anthill with, rule one: "If an AI agent helped write the code, say so in the pull request."
Apache-2.0, still a v0.1 draft.
https://t.co/BN8oYZnRiI
Microsoft lets customers keep their encryption key "in hardware they own and operate".
One part of owning AI. The rest: whose model, whose computers, who can switch it off.
https://t.co/lDuYUVDDN3
A benchmark score says how smart a model is. It says nothing about how much of that model you own.
The AI Ownership Index starts from three questions: "Can it be restricted?" "What happens to your data?" "Will you be locked in?"
Still alpha.
https://t.co/ilEYcwF3Xk
Train it on your own language and data. Local AI gets built by the people who live there.
Mac today, Windows and ChromeOS soon: https://t.co/zDP7pyjSAS
AI ownership gets built for developers. The people who need it most aren’t developers.
A teacher. A small law firm that can’t send client files out. A city working in its own language.
Anthill runs on your own machine, offline. Learns from the answers you approve, no ML degree needed. No token fees, no subscription. Main cost: electricity.
One person on a laptop or a whole organisation in its own cloud, same app.
The UK’s science advisers say the country shouldn’t rely on the big global AI models for everything.
Their advice: decide which AI the UK should own, and test open models it can run on its own computers.
Even government advisers say: don’t rent all of it. https://t.co/Rc220PRj4K
Your own AI, on your own Mac. Not rented.
Anthill runs open models on your computer. Free, AGPL-3.0. Nothing leaves your machine unless you want it to. Your wiki, memory and files stay in an open format you own.
https://t.co/Fx9ONfEBLv
The AI Ownership Index: 21 open models and 9 providers, what you’d actually own, every piece of evidence labelled. https://t.co/3seGcEz5Ng
Full post: https://t.co/mIEN7TswyE
Most people think they own their AI. They own the app.
AI has four layers: weights, compute, data, interface.
Most of us only touch the interface. The one layer that gives you nothing on its own.
“Open” doesn’t settle it. User caps, use restrictions, terms that change overnight. A lease with a nicer label.
If you can’t state your ownership in one sentence from the licence, you don’t have it.
I'm proud and excited alike 🔥
After a successful journey building City AI together with my friend @MikeReiner pushing for the responsible development and application of AI with 400+ ambassadors in 50+ countries that led to the creation of the primer AI conference @WorldSummitAI, the next chapter has started with @onehillai a non-profit organisation with the 💯 goal to enable everyone to OWN THEIR AI.
Why?
Happy to share Anthill 🐜 A new open source project I’ve contributed to!
Your own 100% private, offline AI for knowledge, chats and agents.
Big thanks to @Welsbach_ for starting the initiative and bringing us together.
Check it out: https://t.co/T5l9QLlIGA
Talks about the performance of open-source models is just one side of the coin.
Actual ownership the other: can it be fully used and modified, is it reliable with its output, and are its weights transparent.
Introducing: https://t.co/ilEYcwEw7M
A world where anyone can own, run, and shape their own AI. Where what it knows, and what you teach it, stays yours.
That world doesn't build itself. So today we launch OneHill: a non-profit building real open AI, so anyone can run their own.
Open, or it isn't yours.