Research scientists lead Basis’ efforts to develop a deeper understanding of the conceptual, mathematical, and computational principles of intelligence.
We are looking for people who are technically excellent, and who value probing concepts at their foundations. Our research scientists/engineers aspire to do rigorous, high-quality, robust science, but are not afraid to tinker, make mistakes, and explore radically different ideas in order to get there.
Basis is a collaborative effort, both internally and with our external partners; we are looking for people who enjoy working with others on problems larger than ones they can tackle alone.
Despite the increasing recognition that both having and discovering world models are central to intelligence, current AI systems struggle to replicate this human capability. There remains significant uncertainty about what precisely constitutes a world model, how we might reliably detect if an agent possesses one, and crucially, how we can develop agents that learn these models rapidly and reliably.
Our research within the MARA project aims to develop new foundations and technologies for modeling, abstraction, and reasoning in AI systems. MARA’s overarching goal is to uncover principled methods for how intelligence constructs, refines, and utilizes world models through interactive experimentation. Building these systems will demand advances in knowledge representation, abstraction, reasoning, active learning, reinforcement learning, and a first-principles rethinking of what it means to model the world.
The immediate mission of MARA is to solve concrete challenges such as AutumnBench, physical and simulated robotics benchmarks, and the Abstract Reasoning Corpus (ARC), with the broader mission of building systems capable of learning in an open, growing portfolio of domains using human-comparable amounts of data and interaction.