Ali Jalal-Kamali, Ph.D.
Researcher in behavioral evaluation with focus on AI safety, alignment, and interpretability.
I work on exposing latent behaviors and making them measurable. Given a system (or a dataset) containing behavioral expressions, I explore what can be discovered about hidden behaviors, how early they can be detected, and their impact on the system’s performance. My work expands from geopolitical actors’ relations to human teams’ interactions to frontier language models’ responses. Each study provides a working pipeline along with the findings, so the evaluation systems can be used by others.
Ph.D. in Computer Science, University of Southern California.
Research projects
- TRACE — Divergent Response Modes in Frontier Language Models Under Steering Pressure A symmetric cross-laboratory evaluation of six frontier models, judged blind by one another, with a mechanistic arm on open weights.
- TRIBE — Predicting Team Performance via Communication Behavior Ensembles A domain-independent method for reading team performance from what a team says to itself, early enough to act on.
Links