Hyperbolic geometry, Lisp, and a lab built for agents.
We are still at the beginning of what can be achieved with more intelligence. Through abundant everyday intelligence for everyone. Through accelerating scientific discovery to solve the hardest, most impactful problems like climate change.
We are building to win at the frontier AI. We have our own hyperbolic approach and unique mathematical talent to go beyond the transformer. That gives the direction and focus. What enables us to compete, though, is speed and sustained acceleration, and that comes from our research system.
One of its first components was our own AI-first metrics tracking system. Existing tools were designed for humans watching dashboards whereas ours is built for agents operating independently. The latest unique component is S-expGPU, a whole ML framework. It is essentially PyTorch rethought for AI through Lisp. Lisp enables less code, fewer tokens and fewer errors. It starts fast, runs on NVIDIA today with a natural path to AMD hardware.
Being early in the journey has enabled opinionated choices for going faster than imaginable even a year ago. We are not trying to catch the frontier. We are building to redefine it, for better intelligence.