Ask
Start from a question a product cannot answer yet.
Archymedes Lab
A lab that measures.
The lab studies the questions underneath our products: when an AI should act, whether what it knows is true on the ground, and how to make it cheap enough to run anywhere. We publish the method, the code and the limits.
Method
How the lab works
Measure
Test against ground truth, not against how plausible it looks.
Reject
Drop what fails, however convenient it is.
Publish
Release the paper, the code, the data and the stated limits.
Programmes
Three lines of work
Decisions under uncertainty
When should an AI act on something nobody has tried?
Generative models are good at proposing options and bad at choosing between them. This line of work keeps the two apart: models judge, calibrated maths decides, and an uncertain option is checked against a fallback instead of being vetoed.
Outputs
- Unknown Is Not Risky ↗Paper, October 2026
- cdm ↗Open-source library and data
Grounding in the physical world
Is it true on the ground?
Rebuilding real places as verified, lightweight 3D scenes from free public data. Every layer is registered against the best measurement available and its error is reported; a source that fails against the ground is dropped, however convenient.
Outputs
- geo3dReference build: a mountain valley
- Gisenyi, Lake KivuA second scene, in Rwanda
Efficient intelligence
Can it run on ordinary hardware?
Making capable models cheap enough to run where connections are slow or absent: compression and quantisation for small language models, and tooling that tunes GPU code without trading away correctness.
Outputs
- Small modelsCompression and quantisation research
- WarpLabGPU code tuner, early beta
Publications
Unknown Is Not Risky
Unknown Is Not RiskyCreative Decision Making with Calibrated Decision Models
When should an AI act on an idea nobody has tried? Check it against a fallback instead of vetoing it, and price what the check can miss.
- 01
One check, one number
For a single decision, the value of a check depends on the belief about harm only through its mean. Ignorance and a known coin flip warrant the same action; the difference matters for learning across decisions.
- 02
Checking beats vetoing
On held-out synthetic problems, checking uncertain options halves the regret of vetoing them (0.045 vs. 0.091) at about 1% undetected harm, against 17–23% for acting on them unchecked.
- 03
Tested on a real judge
Over 237 TypeSafe Jev judgments with known ground truth, Jev separates explicit harm from explicit safety almost perfectly (AUROC 0.9998), and cdm has the lowest regret and harm of the policies compared.
- 04
Stated limits
The gain comes from checking, not from the CVaR reliability gate; the Jev benchmark is small and templated; none of this yet shows better decisions on real outcomes.
Open by default
Code, data and limits, in public.
Every result the lab publishes comes with what is needed to check it. Where a benchmark is small or a dataset coarse, the paper says so.