Willy Chung — AMI Labs

Biography

Willy Chung is a PhD candidate in Artificial Intelligence at Sorbonne Université, specializing in world models, machine learning, and high-level reasoning systems. Currently conducting research at Advanced Machine Intelligence (AMILabs), he previously spent over a year and a half at Meta FAIR, where he worked on world models and planning systems to enable AI agents to better understand and interact with complex environments.

Before beginning his PhD, Willy completed a dual-degree program between CentraleSupélec and the Hong Kong University of Science and Technology (HKUST), where he focused on natural language processing and task-oriented dialogue systems. His research interests have since evolved toward building intelligent systems capable of autonomous reasoning, planning, and action in real-world settings.

Alongside his research, Willy has served as a teaching assistant at HKUST, designing and delivering tutorials on deep learning, transformers, and NLP applications. His published work spans multilingual reasoning evaluation, controllable story generation, and data augmentation for reading comprehension systems. Fluent in English, French, and Cantonese, Willy brings an international and interdisciplinary perspective to advancing next-generation AI systems.

Career History

2024-2026

Meta

Research Assitant

Key Papers

A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity

This paper proposes a framework for quantitatively evaluating interactive LLMs such as ChatGPT using publicly available data sets, using 23 data sets covering 8 different common NLP application tasks. We extensively evaluate the multitask, multilingual, and multi-modal aspects of ChatGPT based on these data sets and a newly designed multimodal dataset. We find that ChatGPT outperforms LLMs with zeroshot learning on most tasks and even outperforms fine-tuned models on some tasks. We find that it is better at understanding non-Latin script languages than generating them. It is able to generate multimodal content from textual prompts via an intermediate code generation step. Moreover, we find that ChatGPT is 63.41% accurate on average in 10 different reasoning categories under logical reasoning, non-textual reasoning, and commonsense reasoning, hence making it an unreliable reasoner. ChatGPT suffers from hallucination problems like other LLMs. Finally, the interactive feature of ChatGPT enables human collaboration with the underlying LLM to improve its performance, ie, 8% ROUGE-1 on summarization and 2% ChrF++ on machine translation, in a multi-turn" prompt engineering" fashion. We release a code for evaluation set extraction.

2023 · Association for Computational Linguistics

1,724 citations