Partnerships
Research partnerships, across the sciences
AI systems are increasingly asked to do real scientific work: prove theorems, design molecules, model physical systems, read the literature. We partner with university labs to build the evaluations that tell us whether that work can be trusted, and publish everything that comes out of it openly.
Propose a partnershipWhy partner with us
Published, not proprietary
Everything a partnership produces goes out under an open license, the same as our own work. No internal-only findings, no results that sit in a drawer because they didn't flatter a product.
Small enough to move
We're not routing a proposal through five layers of approval. A partnership usually starts as a conversation with the people who would actually do the work.
Domain expertise leads
We bring evaluation methodology. Your lab brings the domain. The strongest partnerships are the ones where neither side could have built the result alone.
How it works
Domains we're building evaluation in
From mathematics to medicine
Mathematics
Applied math and inverse problems, the modeling work behind rigorous evaluation design.
Physics
Simulation and dynamical systems, the physical sciences AI is increasingly asked to reason about.
Chemistry
Molecular design and property prediction, where a wrong answer is expensive to find out about.
Biology & Medicine
From protein design to clinical reasoning, where evaluation has to meet a much higher bar.
Computer Science
Systems, security, and the software-engineering tasks most benchmarks already try to cover.
Earth & Environmental Science
Planetary and environmental data, where ground truth is hard-won and expensive.
Recent work
A sense of what we publish, and the kind of contribution a partnership is well placed to make.