Portland teen devises method to give tiny AI models more empathy
Seventeen-year-old Henry Xie of Westview High School created a two-step training technique that makes smaller language models respond more empathetically, using outputs from larger models like Gemini and ChatGPT.
Henry Xie, a 17-year-old student at Westview High School in Portland, entered the Regeneron Science Talent Search with a project that seeks to transfer the empathetic capabilities of large language models to faster, cheaper small models. His approach involves two phases: first, providing small models with published empathetic responses from systems such as Gemini and ChatGPT, grounded in psychological theory; second, applying specially crafted prompts that enable the small models to rank stronger empathetic answers higher.
Evaluation showed the trained small models were perceived as more empathetic than their untrained counterparts in at least 90% of instances. Beyond the AI work, Xie heads the school’s computer-science club, competes on the varsity swim team, co-founded the nonprofit Youth for Empathetic AI, and has been a three-time semifinalist in the CyberPatriot National Youth Cyber Defense Competition.
Why it matters
Making empathetic AI accessible on low-power devices could improve user experience across everyday apps.
In this story
