Independent AI research lab

Building machine intelligence we can reason about.

We study how large models learn, fail, and generalize — and use what we find to build systems that are capable, interpretable, and safe to rely on.

Research areas

Three threads run through our work, from foundations to deployment.

Foundations

Rethinking how models represent, retrieve, and reason over information at scale.

Alignment

Opening the black box — tracing what a model has learned and why it acts the way it does.

Deployment

Rigorous testing methods that catch failure modes before they reach the real world.

Small teams, long timelines, published results.

Researchers work in pods of three to five, owning a problem end to end.

Findings are written up and released — good results and dead ends alike.

We fund research in multi-year arcs, not quarterly deliverables.