Xi(Sean) Chen 陈熙

Hi! I'm an Applied Scientist at the Pretraining Integration & Scaling Team, Amazon AGI Foundations — where I work on pre-training and mid-training for Amazon Nova series of LLMs in terms of model design, training dynamics, data mixture optimization, synthetic generation, and evaluation.

Before joining Amazon, I had PhD study in computer science at University of Massachusetts Amherst (UMass), where I am luckily advised by Przemyslaw Grabowicz. I also work closely with Brendan O'Connor (UMass), Ethan Zuckerman (MIT&UMass), Scott Hale (Oxford), and David Jurgens (UMich). Before pursuing PhD, I led research on evolving graphs at Z.AI (Chinese OpenAI, developing GLM series of LLMs) under the guidance of Jie Tang (Tsinghua).

My research is pioneering event evolution within textual media graph at global scale , with interdisciplinary study of LLM training, machine learning, information retrieval, natural language processing, network science, and social science.

My curiosity expands to philosophy, psychology, fine arts, economy, and politics. In leisure time, I enjoy playing tennis and strategy games. I ranked global top50 of Autochess (50/millions, a world-class strategy game for playing DOTA on chess table).

Highlights:

• We released Nova LLM series Nova 2 Pro, Nova 2 Lite, and Nova 2 Sonic! Check evaluation by Artificial Analysis [Post], [Tweet]
• From trillions of cross-lingual news similarities, we found economic factors dominate the rhythm of global media coverage!
• We found a terrorist attack in Africa need 20x more deaths for same media attention as one in Eastern Enrope! [Masslive Coverage]