Who Owns Your AI? Mira Murati Wants It To Be You
Imagine lending your favourite hoodie to a massive corporation, and then being told you can never get it back — and also, the hoodie now makes all your decisions. That is roughly what centralised AI ownership feels like, and Mira Murati’s Thinking Machines Lab has had enough of it. Their new essay, “The Future Worth Building Is Human,” makes a bold technical argument: AI should belong to the people actually using it, not just the companies building it.
So What Is Thinking Machines Lab Actually Saying?
The essay frames three big ideas as genuine technical challenges rather than just feel-good philosophy. Think of it like a school science project, except instead of a baking soda volcano, the stakes are the entire future of human civilisation. No pressure.
- Human participation: Real people should be actively involved in shaping how AI behaves, not just clicking “I agree” on a terms-of-service document nobody reads.
- Model ownership: Teams and individuals should be able to own their own AI models — like owning a car instead of forever renting one from a company that controls where you can drive.
- Decentralised alignment: AI values and behaviour should not be set by a single organisation sitting in a boardroom. Instead, alignment should spread out across many different groups with their own needs and contexts.
What On Earth Are Model Weights?
Great question! Imagine your brain as an AI model. Every experience, lesson, and embarrassing memory you have ever collected has shaped tiny connections inside it. In an AI, those connections are called model weights — they are essentially the numbers that determine how the AI thinks and responds.
When a company trains an AI, they set those weights. If you cannot touch or customise those weights, you are basically borrowing someone else’s brain and hoping it agrees with your values. Thinking Machines Lab argues that teams should be able to train and keep their own weights, making the AI genuinely theirs.
Enter Tinker and the Magic of LoRA Fine-Tuning
This is where things get excitingly nerdy. Thinking Machines Lab has developed a tool called Tinker, which uses a technique called LoRA fine-tuning. LoRA stands for Low-Rank Adaptation, which sounds complicated but is actually quite clever.
Think of a massive pre-trained AI model like a giant cookbook with thousands of recipes. LoRA fine-tuning lets you add your own personalised sticky notes throughout that cookbook without rewriting the whole thing. You keep the original recipes but your custom notes change how you actually cook. The result? An AI model that reflects your specific needs, your team’s knowledge, and your context — without costing you a fortune to retrain from scratch.
Why Does This Actually Matter?
- A hospital team could fine-tune their AI to understand their specific medical protocols without sharing sensitive data with a third party.
- A small business could customise AI behaviour to match their brand voice instead of sounding like a generic chatbot with the personality of a cardboard box.
- Researchers could experiment freely without waiting for a big tech company to approve their ideas.
The Bigger Picture: Decentralised Alignment
One of the most interesting arguments in the essay is about alignment — making sure AI does what humans actually want. Right now, most AI alignment decisions happen centrally, meaning one organisation decides what is good for everyone. Thinking Machines Lab believes this is risky, like letting one person control every traffic light in every city on Earth.
By giving teams ownership of their own model weights, alignment naturally becomes decentralised. Different groups can shape AI behaviour according to their own communities, cultures, and values. It is a genuinely fresh way of thinking about one of AI’s hardest problems.
The Bottom Line
Thinking Machines Lab is not just writing a nice essay about feelings. They are making a concrete technical argument backed by real tools like Tinker. If they are right, the future of AI looks less like a giant vending machine controlled by a few companies and more like a garden where everyone gets to grow something of their own.
And honestly? That sounds like a future worth building.