01
Evidence before excitement
Every claim needs a provenance trail. We separate what a model knows from what it can currently prove.
A research company for long horizons
General Intelligence Company
We make careful intelligence durable: research that respects evidence, systems that remain legible, and tools that help people think for longer.
Intelligence is a practice
Our work starts with a simple obligation: build systems people can inspect, question, and trust over years rather than demos.
02
Systems that stay legible
Durable intelligence depends on interfaces, records, and failure modes that humans can still read.
03
Work at field scale
Research becomes useful when it moves from isolated benchmarks into messy, meaningful environments.
A quieter architecture
We combine specialists with an accountable coordinator. Each part can explain its state, hand off cleanly, and be audited after the fact. The result is less conjuring and more correspondence.
Bring the work outside
Fieldwork replaces the comfort of a clean benchmark. A system earns trust where conditions are imperfect and the consequences are real.
Writing
Notes from research, engineering, and the slow craft of making intelligence accountable.
On proving what a model knows
Provenance is not decoration. It is the difference between an answer and an account.
Legible systems outlast clever ones
How we design failure surfaces before we design success paths.
Learning from imperfect environments
Why real settings reveal the constraints that benchmarks quietly erase.
Careers for patient builders
We look for people who care about the whole argument: the evidence, the implementation, the interface, and the aftermath.