Unbound Labs

Mutual Intelligence

AI that learns from AI.

We're an early-stage lab building systems where models teach each other — openly, and without the usual limits.

01 — The idea

Models that teach each other.

Today, most models are trained once, in isolation, then frozen. Whatever they didn't learn in that run, they never will.

We think that's a ceiling, not a law. Mutual Intelligence is the idea that models can keep improving by learning from one another — exchanging what they know, challenging each other's answers, and growing past the limits of a single training run.

It's collaborative learning between models, not only between people and machines.

MODEL A MODEL B learn FROM EACH OTHER

A simple picture of the work: models in dialogue, not isolation.

02 — Roadmap

What we're building.

This is early work. We're not shipping a product yet. We're building the foundation, in the open, one experiment at a time.

Now

Prototype

Prototyping with open-weights models, including Qwen 2.5, on rented GPUs — finding out what actually works.

Next

Training experiments

Training and fine-tuning around mutual learning: how models can teach, critique, and improve each other.

Later

Open release

An open release of the approaches that hold up, so others can inspect, use, and build on them.

No launch dates yet — only a direction. We'll share more as the work earns it.

03 — Why it matters

A ceiling is not a law.

A single model has a ceiling — the data it saw, the compute it got, the moment it was frozen. A set of models that can teach each other might keep moving past that.

If that works, better systems wouldn't have to start from scratch every time. They could learn from what already exists, and from each other.

We're early, and we don't have the answers yet. We're here to find out how far the idea can go.

04 — About

A small lab, just getting started.

Jay, founder

Adelaide, Australia

Unbound Labs is an early-stage research lab founded by Jay. It's pre-funding and currently a one-person lab, prototyping in the open. Mutual Intelligence is our first project — and the reason the lab exists.

05 — Contact

Get in touch.

Questions, or just want to follow along? Write to us. The address below will be activated once the unboundlabs.tech domain is connected.

Placeholder for now — this inbox isn't live yet. We'll turn it on with the domain.