Henry Xie
Westview High School
Portland, OR
Distilling Empathy From Large Language Models
Henry Xie, 17, of Portland, developed a way to transfer empathy from large AI models to smaller models for his Regeneron Science Talent Search computer science project.
View Poster
Large language models (LLMs) can respond to human emotions in caring ways, but require significant computing power. Small language models (SLMs) — used in smartphones and many everyday applications — are faster and cheaper, but often less empathetic. Henry built a two-step model for training SLMs to display more empathy.
First, he fed them published examples of empathetic responses from LLMs such as Gemini and ChatGPT, grounded in psychological theory. Then he used targeted prompts to help SLMs distinguish stronger empathetic responses from weaker ones. Henry used LLMs to judge their responses. Compared to untrained SLMs, the trained SLMs were seen as more empathetic in at least 90% of cases. His approach could make empathetic AI more widely available.
The child of Fei Xie and Huaiyu Liu, Henry attends Westview High School, where he leads the computer science club and the varsity swim team.
Beyond the Project
Henry is co-founder and president of Youth for Empathetic AI, a nonprofit that brings together students and researchers to create ethical, human-centered AI.
FUN FACTS: Henry was a three-time semifinalist in the CyberPatriot National Youth Cyber Defense Competition, one of the world’s largest youth cybersecurity contests.
