New Theory is developing scalable foundational models using Geometric Deep Learning to overcome the performance limitations of transformer-based architectures. This technology enables significant performance improvements across Robotics, Generative AI, Language, Reasoning, and Scientific Discovery.
Funding
Funding not disclosed
Founders
Product
Problem
Transformer-based AI models require massive datasets and computational resources, leading to high costs, energy consumption, and limitations in scalability. These models often struggle with generalization and explainability, hindering their applicability in robotics, autonomous vehicles, and scientific discovery.
Solution
New Theory is developing AI architectures based on geometric deep learning and principles from neuroscience to overcome the limitations of traditional transformer models. Their approach focuses on building structured world models that exhibit strong generalization with significantly less data and compute. By leveraging mathematical primitives evolved in natural intelligence, New Theory aims to create AI systems that are faster, more efficient, and more explainable. These next-generation models are designed to unlock advancements in robotics, autonomous systems, generative AI, and scientific understanding.
Target Audience
The primary audience includes researchers and developers in robotics, autonomous vehicles, generative AI, and scientific discovery seeking more efficient and scalable AI solutions.
Features
- Geometric deep learning architectures for improved scaling compared to transformer-based models
- Multi-modal world models trained with a fraction of the data and compute
- Models designed for perception, generation, planning, semantic understanding, and reasoning
- Integration of mathematical primitives inspired by neuroscience