Kelvin Xu

I am a researcher interested in large language model pre-training at Google DeepMind. I most recently led scaling laws work that became part of the Gemini 1.5 model family. I previously worked on Gemini 1, PaLM-2 and trained some of the first preference models used in Project Bard (now known also as Gemini). I am interested in all things related to creating useful AI systems.

I did my PhD at UC Berkeley in the EECS department where I was fortunate to be advised by Prof. Sergey Levine. Before starting at Berkeley, I had the pleasure of working at Google as part of the inaugural Brain Residency Program (now AI Residency).

I did my MSc at the MILA lab at the Université de Montréal under the supervision of (Turing Award Winner) Prof. Yoshua Bengio and Prof. Aaron Courville. I also worked closely with Prof. Kyunghyun Cho. Prior to that, I did my B.A.Sc in the Engineering Science Program at the University of Toronto where it all started.

email / google scholar
(website template credits)