Making ASI for the Real World
We’re building new AI architectures that break neural scaling laws—making embodied and generative intelligence fast, efficient, and explainable.
Our Vision
Scale-is-all-you-need doesn’t scale
Solution
We see a future where advanced intelligence is seamlessly integrated into daily life—powering robotics, autonomous vehicles, generative media, and scientific discovery. But today’s transformer-based models are power-hungry, data-intensive, and hard to control—limiting their ability to scale into real-world, embodied intelligence.
Scaling today’s models requires power-plant-level energy, hundreds of millions of dollars, and months of training. Recent results now cast doubt on whether scaling laws will continue to hold at all.
We need a new path to intelligence.
We’re building next-generation frontier AI built on new mathematical principles discovered in neuroscience and geometric deep learning. These structures unlock orders of magnitude better scaling than the transformer-based architectures and enable a future of embodied intelligence.
0 / 0
Our Vision
We see a future where advanced intelligence is seamlessly integrated into daily life—powering robotics, autonomous vehicles, generative media, and scientific discovery. But today’s transformer-based models are power-hungry, data-intensive, and hard to control—limiting their ability to scale into real-world, embodied intelligence.
Scale-is-all-you-need doesn’t scale
Scaling today’s models requires power-plant-level energy, hundreds of millions of dollars, and months of training. Recent results now cast doubt on whether scaling laws will continue to hold at all.We need a new path to intelligence.
Solution
We’re building next-generation frontier AI built on new mathematical principles discovered in neuroscience and geometric deep learning. These structures unlock orders of magnitude better scaling than the transformer-based architectures and enable a future of embodied intelligence.
New structures for perception and cognition
Natural Intelligence
Brains have evolved mathematical primitives for representing the world—efficiently discovering, representing, and reasoning about its underlying structures.
World Models
Structured world models exhibit strong generalization far out of their training distributions, forming a robust foundation for perception, generation, planning, semantic understanding, and reasoning.
We’re building it
New Theory is building structured, general-purpose architectures for multi-modal world models with compositional generalization, achieving SOTA when trained with a tiny fraction of the data and compute.
