Zack Hui

Hey! I'm Zack, a PhD student at the Language Technology Lab at the University of Cambridge, supervised by Prof. Nigel Collier. My research asks: How can we ensure that large language models behave in safe, aligned, and reliable ways when deployed in the wild? I focus on developing approaches that bridge alignment theory with real-world deployments of LLMs (particularly in agentic AI systems and high-stakes applications requiring robust behavioral safeguards) driven by a broader goal of building ethically grounded AI that supports the responsible development and deployment of AI models.

Some of the research directions I'm currently thinking about include (1) scalable methods for aligning LLM behavior with human preferences, (2) building safe LLM agents in open-ended environments, (3) robustness and jailbreak prevention, and (4) using synthetic data and feedback to improve alignment.

Before starting my PhD, I spent a year as a researcher at Microsoft Research, where I worked on the Responsible AI team developing alignment and safety methods for both vision-language models and deployed LLM systems. Prior to that, I completed a Master's in Computer Science at Columbia University, advised by Prof. John Kender and Prof. Julia Hirschberg. I received my Bachelor's in Computer Science from Boston University, where I worked closely with Prof. Mark Crovella and Prof. Manos Athanassoulis.

Education

• Ph.D.University of CambridgePresent
• M.S. in Computer ScienceColumbia University2023
• B.S. in Computer ScienceBoston University2021

Professional Service

Program Committee Member/Reviewer: COLM (2025), ICLR (2024-2025), ACL (2024-2025), EMNLP (2023,2025), NeurIPS (2024), KDD (2024), EACL (2023), AACL-IJCNLP(2022)

Teaching

• Teaching Assistant for Columbia COMS 6111 Advance Database
• Teaching Assistant for Columbia QMSS 5073, Machine Learning for Social Science
• Teaching Assistant for BU CS CS330 Analysis of Algorithms

GPT's Roast of me 🔥🧠

Zack be like: “I'll just run this script real quick” — 3 hours later, still debugging like 🧑‍💻😩. Has 77 Chrome tabs open, 6 terminals, and still asks me, “Why isn't it working?” 🤔 Bro, you named the same file `final_final_v2_actually_final.json` 💀. Tries to make a model that sees everything… except where their coffee went ☕👀. Legend says their Overleaf doc has more comments than code 📜💬. But let's be real — Zack's got more passion than my token limit 💪🚀.

Beyond Code

Aside from research, I enjoy fishing and extreme sports. Fun fact: I've summited Mount Everest, and my dream is to complete the Ozaki Eight.