Andrea Cui hi there!

Cognitive scientist · AI researcher

Andrea Cui

CogSci PhD, currently doing GenAI model evals at Luma AI.

I'm a cognitive scientist with a PhD in Cognitive Science and master's degrees in Computer Science and Statistics, working on evaluation, post-training, and human preference alignment for agentic image and video generation at Luma AI. Beyond that, I explore automated research for self-evolving world models, along with video-physics evaluation.

Work

Luma AI

Now

GenAI Model Evaluation

Evaluation, post-training, and human preference alignment for agentic image and video generation, plus automated research for self-evolving world models and video-physics evaluation.

Rockstar Games

Fall 2025

Research Data Scientist · Internship

A research-oriented data science role developing theory-grounded behavioral models, machine learning methods, and causal frameworks.

University of Illinois Urbana-Champaign

PhD

PhD in Cognitive Science · M.S. in Computer Science · M.S. in Statistics

Researcher at the intersection of human cognition, decision-making, and computational modeling.

Publications

Selected research on human cognition, AI evaluation, and human–AI interaction.

Full list on Google Scholar ↗

Empowered XR through Generative AI: Balancing Superpowers and Risks

CHI 2026

Analyzes 135 recent works to map the “superpowers” generative AI grants users in immersive spaces, along with the risks they carry, distilling both into a taxonomy and design guidelines.

View paper ↗

LLM Integration in Extended Reality: A Comprehensive Review of Current Trends, Challenges, and Future Perspectives

CHI 2025

A comprehensive review of how large language models are being woven into extended reality, mapping current trends, open challenges, and where immersive AI is headed next.

View paper ↗

RPGBench: Evaluating Large Language Models as Role-Playing Game Engines

arXiv, 2025

A benchmark that tests whether LLMs can run a full role-playing game, keeping state, enforcing rules, and sustaining a coherent interactive story.

View paper ↗

EyeSee: Enhancing Art Appreciation through Anthropomorphic Interpretations from Multiple Perspectives

CHI 2025

An interactive system that lets artworks “speak” from multiple anthropomorphic perspectives, opening new ways for viewers to connect with art.

View paper ↗

Do Language Models Have Bayesian Brains? Distinguishing Stochastic and Deterministic Decision Patterns within Large Language Models

NeurIPS 2024 Workshop on Behavioral Machine Learning

Probes whether LLMs decide like Bayesian reasoners, teasing apart the stochastic and deterministic patterns hiding inside model behavior.

View paper ↗

Complex Background Information Slows Down Parallel Search Efficiency by Reducing the Strength of Interitem Interactions

Journal of Experimental Psychology: Human Perception and Performance, 2023

Shows that cluttered backgrounds slow human visual search not by adding noise, but by weakening the interactions between the items we search through.

View paper ↗

Order Constrained Analyses of Eyewitness Memory Accuracy

Vision Sciences Society (VSS), 2024

Uses order-constrained modeling to test which commonly reported patterns among accuracy, confidence, and response time genuinely hold in eyewitness memory data.

View paper ↗

Blog

Essays on AI, cognition, and creativity.

Also on Substack ↗ Also on Medium ↗