What is it?
Trim is building an AI model that can simulate real-world physical systems evolving over time. For example, given the starting position of waves on a beach, the model generates how those waves move forward in time.
What’s wrong with a traditional physics simulation?
Traditional physics simulations take exponentially longer to run as you simulate more dimensions and polynomially longer as you increase the size of the simulation. The Trim Transformer’s linear-attention scales linearly in computation time with respect to both dimensions and grid size. Also, if you want to simulate twice as far into the future, it currently takes twice as long. Our architecture takes logarithmically longer.
Combining these architectural advantages, latency sensitive tasks like an autonomous vehicle choosing its path drop by several orders of magnitude and previously computationally infeasible tasks like detecting gravitational waves become possible.
How does it work?
Trim trains our models by running traditional physics simulations and feeding the results into our pipeline. The Trim Transformer is a custom implementation of Galerkin-type attention. One way to think of our models is as a constant-time lossy lookup table.
We will be open-sourcing the Trim Transformer and our first models in the coming weeks. A series of blog posts will explain our architecture and the design challenges we’ve overcome.
How will Trim solve quantum gravity?
Even the strongest gravitational waves are extremely weak. Two super-massive black holes colliding generate vibrations measured by LIGO that are about as powerful as a pickup truck turning on one mile away. The only feasible way to extract the data out of the noise is by looking in the sky at an event, simulating what the waves should look like, and picking the needle out of the haystack on a years-long timeline.
The problem plaguing astronomers for the last decade is that gravitational wave frequencies that are quick to simulate aren’t very helpful for finding flaws in general relativity. The frequencies of interest would take thousands of years to simulate and identify in all the noise. A new gravitational wave detector, LISA, will soon be launched into space, capable of detecting new frequencies that could expose hints of new physics. Trim’s model makes the task of finding that data within reach for the first time.