Patrick Zhao

I am an MSc thesis student at Simon Fraser University, advised by Dr. Nicholas Vincent.

I study how large language models can serve as tools for social science, focusing on methods for multi-agent LLM simulations. I also run Peasant Labs, where our mission is to create open-source infrastructure for people to collectively govern the data that trains AI, and accelerate the precedent of ethical data labour standards.

Recent News

Model Capacity Frontier: A Development Framework for LLM-Supported Social Simulation

arXiv preprint

Patrick Zhao, Yuxuan Li, Hirokazu Shirado

Mechanism Plausibility in Generative Agent-Based Modeling

FAccT 2026

Patrick Zhao, David Huu Pham, Nicholas Vincent

Mechanism Plausibility in Generative Agent-Based Modeling

PoliSim @ CHI 2026 Best Paper Nominee Best Paper Nominee at PoliSim @ CHI 2026

Patrick Zhao, David Huu Pham, Nicholas Vincent

Empowering Users Together: Connecting Algorithmic Collective Action and Explainable AI

Algorithmic Collective Action Workshop @ NeurIPS 2025

Ayana Hussain, Cole Michael Thacker, Patrick Zhao, Aditya Karan, Nicholas Vincent

Let Them Down Easy! Contextual Effects of LLM Guardrails on User Perceptions and Preferences

EMNLP Findings 2025 News Coverage Mentioned in Forbes — Aug 2025

Mingqian Zheng, Wenjia Hu, Patrick Zhao, Motahhare Eslami, Jena D. Hwang, Faeze Brahman, Carolyn Rose, Maarten Sap

An Audit and Analysis of LLM-Assisted Health Misinformation Jailbreaks Against LLMs

AIES 2025

Ayana Hussain, Patrick Zhao, Nicholas Vincent

All publications →

Why are you including Pangram scores?

I want to be transparent with what you, the reader, will be consuming! I'd rather readers not have a visceral reaction of feeling misled later on after reading or testing it themselves. You may see varying metrics across my papers; my research is a collaborative process with different people (all super cool) that don't always use AI-assisted writing to the same degree.

Note: Some papers include LLM outputs as research data, such as generated prompts or responses in an appendix, so those count toward the score.

Last update: Pangram 4.1, Nov 2026