Research · Oct 02, 2026 · 14 min
Towards reliable multi-step reasoning under uncertainty
Our working agenda for models that check their own reasoning, recover from errors, and know when they don't know.
Research
Sharing what we learn, as we learn it.
Safety · Sep 18, 2026 · 11 min
Interpretability as an engineering discipline
Why understanding internal computation must become as routine as testing — and what tooling it will take.
Infrastructure · Aug 27, 2026 · 9 min
Designing compute for long-horizon research
Notes on the systems we're building to run experiments that last weeks, not hours.
E
Essay · Aug 06, 2026 · 8 min
What we mean by general intelligence
A precise definition matters. Ours centers on transfer, adaptation, and competence in unfamiliar domains.
E
Evaluation · Jul 15, 2026 · 12 min
Measuring capability without measuring memorization
Principles for evaluations that remain meaningful as models scale and training data grows.
R
Research · Jun 24, 2026 · 10 min
Continual learning without catastrophic forgetting
An overview of the open problems in systems that keep learning after deployment.
P
Policy · Jun 03, 2026 · 6 min
Our commitments on responsible release
How we decide what to publish, what to withhold, and how we stage access to new capabilities.
E
Essay · May 12, 2026 · 5 min
Why Koodos is a nonprofit
The structure of an institution shapes what it optimizes for. We chose ours deliberately.