Trim

Trim is building an AI model designed to simulate real-world physical systems as they evolve over time. Traditional physics simulations become exponentially slower as more dimensions are added and polynomially slower as the simulation grid grows, making many tasks computationally impractical. Trim addresses this by using a Transformer architecture with linear-attention, which scales linearly in computation time with respect to both dimensions and grid size, enabling faster and more feasible simulations.

The core of the product is the Trim Transformer, which uses a custom implementation of Galerkin-type attention. The model is trained by running traditional physics simulations and feeding the results into its pipeline, effectively functioning as a constant-time lossy lookup table. This approach allows for latency-sensitive applications, such as autonomous vehicle path planning, to run significantly faster, and makes previously infeasible tasks, like detecting gravitational waves, possible.

The product is aimed at industries and researchers who need to simulate complex physical systems quickly, such as autonomous driving and gravitational wave detection. The website does not specify pricing, packaging, or trial options.