Our science
Does the AI Actually Learn Biology?
[FIXXXX]How to assees to what extent AI actually learn biology? Frontier models are moving toward world models that understand the consequences of their actions. We expect that within ten years, and we build the measurement layer that tells you which models are getting there — so your teams adopt them early, on evidence, and ahead of your competitors.
Platform
One layer between model and decision.
Our platform aims at innovating measurement and metrics to evaluate agentic AI systems and frontier models, providing scalable methods to annotate your proprietary data and make it AI-ready, compatible with any AI vendor.
Why Generalizability Metrics Are Needed
A model can use biology without having learned it.
Our science focuses on explainability and generalizability of intelligent systems, toward decoding biological complex systems by distillation and encoding them to AI. Our science similarly views AI as a complex system, and tests whether the AI actually learns biology.
Accuracy on a held-out split says a model fits the data it was given. Generalizability metrics ask a different question: does the model carry an understanding that survives a new target, a new assay, a new patient population — the conditions your next decision will actually be made in.