Ekin Dogus Cubuk runs a startup to accelerate physics R&D using AI

“What can physicists do? ” is an interview series that profiles physicists who opted for careers outside of academia.

What is your area of research?

Developing machine-learning algorithms and simulation tools for studying amorphous and crystalline solids—glasses, battery materials, and topological insulators are examples.

What attracted you to industry?

I thought I’d stay in academia. I wanted to do research. But I saw professors spending so much time writing grant proposals, and I am not a good writer.

My best friend from undergraduate told me he was able to do more theoretical physics at Google than in academia, and working for industry meant there was no grant pressure. I first had a one-year residency position with Google Brain and then was hired in 2018.

What did you do at Google?

I did machine learning for physics, and some things that were more targeted for Google, like modeling of neural networks and augmentation algorithms, which are used to improve deep learning for images. I was there for eight years and was leading a group of 22 people.

Last year, I left to cofound a startup company, Periodic Labs.

What does Periodic Labs do, and why start your own company?

Periodic Labs aims to bring together large language models and theoretical and experimental physics, and to use LLMs to accelerate physics R&D. One of our goals is to discover new superconductors.

I wanted to start my own company because experiments are extremely important for solid-state physics; theory and simulations are not enough. And it was proving difficult to build a lab in a big software company. Another reason was that LLM researchers and physicists don’t usually work together, and that was crucial for our aims.

What do you like about your work?

In the startup, everyone is focused on one goal. And we—chemists, physicists, LLM experts—learn a lot from each other.

What nonphysics skills did you need to learn?

Writing scalable and efficient code is very important. And I’ve had to learn to think about the potential financial impact of my research.