Kai Xu 许凯
About
Kai Xu

I'm a second-year PhD student at New York University advised by He He. My ongoing research is in developing AI evaluations and safety training methods.

Previously

I completed my undergraduate degree at Yale University, conducting research in computational social science with Nicholas Christakis and bioclimatic modeling with William Lauenroth. In summer 2024, I did research in species distribution modeling with Nash Rochman at the National Institutes of Health. Outside of research, I enjoy composing, learning about human behavior, and cooking up the occasional side project.

Research

In a world where AI is becoming increasingly pervasive, I plan to explore more of these questions during my research career:

How do we make AI more good?

I define "good" as being beneficial for physical, mental, and social health, and "bad" as detrimental to these ends. I'm especially drawn toward scientific applications as a key motivator.

How do we make AI less bad?

In this direction, I'm interested in developing benchmarks that shed light on failure modes, as well as safety training methods that improve reliability and robustness.

Papers
PaintBench: Deterministic Evaluation of Precise Visual Editing

Kai Xu, Ellis Brown, Shrikar Madhu, Rob Fergus, He He, Saining Xie