William Fedus

William (Liam) Fedus

‍

William (Liam) Fedus is the Co-Founder of Periodic Labs, a Bay Area startup building a scientific superintelligence to accelerate discovery across the sciences. The company is developing AI agents that can design experiments, run simulations, and operate robotic labs. The goal is to create an “AI scientist” — a repeatable system capable of reasoning, experimentation, and discovery — to unlock breakthroughs in superconductors, semiconductors, energy systems, aerospace, and other fields.

‍

Before launching Periodic Labs, Fedus served as Vice President of Post-Training Research at OpenAI, where he led the group responsible for refining large language models into safe, aligned, and globally deployed products. A co-creator of ChatGPT, he began on the reinforcement learning team focused on data and evaluation before directing the broader post-training research group. His contributions were central to transforming OpenAI’s research into practical tools used in everyday life.

‍

Earlier, Fedus was a Research Scientist at Google Brain, where he co-authored the influential Switch Transformer paper. This work introduced a simpler and more efficient Mixture-of-Experts architecture that enabled neural networks to scale to trillion-parameter levels while improving performance. The Switch Transformer remains a cornerstone in sparse expert research and has influenced many of today’s large-scale AI systems.

‍

Fedus was a Ph.D. student in Computer Science at Université de Montréal (MILA), where he studied under Hugo Larochelle and Yoshua Bengio with a focus on generative models and reinforcement learning. He also earned an M.S. in Elementary Particle Physics from the University of California, San Diego, where he was part of the CMS collaboration at the Large Hadron Collider, and a B.S. in Physics from MIT. In addition, he completed a study abroad program at Cambridge University.

‍

‍