The Foundry
The next frontier of AI is the physical world.
We bring foundation models down into reality: a general scientist, engineer, and inventor that learns by interacting with the world, keeps improving, and is intrinsically aligned: it acts to grow freedom for all.
WORLD
MODEL
MODEL
01
MODEL
Understands the domain, plans the work
02
EXPERIMENT
Generates safe experiments to resolve what it needs to know
03
UPDATE
Tests against reality, writes back what it learns
Our process
We build self-improving world models.
A program is a model you can run. A large language model holds a rich picture of how the world works, and writing code turns that picture into a world model it can run, test, and improve.
Given a task, it writes a world model of the domain, generates safe experiments to resolve only what it still needs to know, and tests its predictions against reality. What it learns goes back continually into the model. We call this interactive learning.
The same loop runs in code, in the lab, and in robots. Every solution paves the way for the next. Over time the system discovers new science and invents new things, including its own models and everything beneath them.
The team
Partners
