Ziwei Gu

Hi, I'm Ziwei.

I am a second-year Ph.D. student in Computer Science at Harvard University, advised by Dr. Elena Glassman. Currently, my research focuses on augmenting human cognition and efficiency by leveraging large language models (LLMs) and interactive techniques, aiming to ensure that potential AI errors can be easily noticed, judged, and recovered from, a concept I formalize as AI-resiliency.

Before coming to Harvard, I graduated with a bachelors in Mathematics and Computer Science (December 2020) and masters in Computer Science (May 2021) from Cornell University.

How can we get a better understanding of the mechanism through which an LLM-based reading assistance tool (GP-TSM) supports reading? We conducted an eye-tracking user study with 24 participants, followed by an analysis of the unique gaze patterns associated with GP-TSM.

We propose the idea of "AI-resilience" and an LLM-powered technique that supports reading through recursive summarization while allowing readers to easily notice and recover from LLM summaries they disagree with.

Large language models (LLMs) are capable of generating multiple responses to a single prompt, yet little effort has been expended to help people make use of this capability. In this paper, we explore how to present many LLM responses at once.

Interaction logs can be extremely complex yet useful. Breaking down event logs into goal-directed segments can make it easier to understand user workflow.

We ask a fundamental research question: How do core design elements of debiasing systems shape how people reason about biases? We present distinctive sensemaking patterns and surprising findings from think-aloud studies.

We present Silva, an interactive tool that utilizes a causal graph linked with quantitative metrics to help people find and reason about sources of biases in datasets and machine learning models.

We trained a deep learning model (Transformer) for open information extraction, modeled as a sequence to sequence transduction task. We showed that our model was competitive with state-of-the-art systems, but without the dependencies on other NLP tools.

Layout inspired by this template