Patrick Chao

Hello! I am a final-year Statistics Ph.D. student at the Wharton school, University of Pennsylvania, gratefully advised by Edgar Dobriban. I am broadly interested in understanding large models and AI safety. Currently, my research is on LLMs, diffusion models, and adversarial attacks.

For my undergraduate education, I graduated from UC Berkeley, receiving Bachelor's degrees in Computer Science, Mathematics, and Statistics with honors. I was very fortunate to have been advised by Will Fithian and Horia Mania.

Previously, I have interned at Amazon AI, working on diffusion models for causality, and at Jane Street Capital as a trading intern.

Email  /  Github

We propose PAIR, an automated method that uses a language model to systematically generate semantic jailbreaks for other language models, often in under twenty queries. PAIR is more computationally efficient than state-of-the-art methods by many orders of magnitude and only requires black-box access.

When estimating an average under distribution shifts, folk wisdom is to use something robust like the median. Surprisingly, under bounded (Wasserstein) shifts, we show that the sample mean remains optimal. For linear regression, we show that ordinary least squares remains optimal.

Systematically finding adversarial prompts for generative image models like Stable Diffusion and text models like GPT.

Using diffusion models to answer interventional and counterfactual queries by modeling observational data, achieving state-of-the-art performance.

A powerful and robust multiple testing method using covariates to model the local false discovery rate by fitting a Gaussian mixture model.

We show the standard definitions of conic sections are not equivalent in hyperbolic geometry and we define generalized versions.

Last updated October 2023.
Website template from Jon Barron.