Researching
autonomous systems for
superhuman mathematical reasoning

OUR MISSION

Our mission is to discover emergent capabilities in autonomous systems designed to rapidly advance toward superhuman mathematical reasoning. We believe the foundational challenge lies in autonomous formalization and program synthesis—cracking this enables systems that can improve themselves.

APPROACH

Our effort lies at the intersection of cutting-edge reinforcement learning and advanced reasoning, structured within a purpose-built scaffolding. Our core research approach for achieving superhuman mathematical reasoning is to enhance the autonomous capabilities of language models through reinforcement learning—equipping them with precisely the tools and environments they require, and scaffolding them toward multi-system.