We're a research org + grantmaker focussed on the science of evals and on spiky processes.
We look for evidence for and against near-term transformative AI and track developments in AI economics, capabilities, safety, and politics in real-time. Sign up here.
We make grants with an implicit spec described here. We're most interested in prospects for narrow, or non-recursive, or human-augmenting AI.
Why?
We don't know what's going on. And we want better evidence than current AI evaluations can give us.
For various reasons (selection, groupthink, ethos, incentives), independent AI research remains undersupplied and disproportionately powerful on a per-dollar basis.
AIs are tools, agents, and maybe soon a third thing (an autonomous competitor beyond the speed and potency that current social systems are designed to handle). We want to buy time for society to reckon with the first two.
This could happen by promoting "centaur" tools that lift human marginal product and extend human capabilities, or by preventing dangerous recursive spikes and overhangs in AI development.
Who?
Co-founder of the research consultancy Arb and fellow at Cosmos and Foresight. PhD from Bristol; past lives in epidemiology, metascience and AI.
Formerly led much of sales and partnerships at Google Cloud. In early 2020 he founded a real-time Covid tracking org and now leads Ergo Impact, a grantmaker.
Writer at the intersection of machine learning, philosophy, and literature. PhD from Harvard; essays in Aeon, Glass Bead and The Gradient.
Foresight fellow and lead author on the State of Brain Emulation Report 2025. Co-founded OpenBioML.
Co-author of the annual Shallow Review of Technical AI Safety, mapping the field’s active research agendas.