HKU IDS Scholar Seminar Series #28:
LLM Math: Prospects and Challenges
Abstract
This project examines the striking achievements and unique challenges of large language models’ (LLMs’) mathematical and logical (collectively, “formal”) reasoning performance. We observe that training LLMs to excel at formal reasoning requires clearing a significant hurdle: namely, we require a predominantly inductive inference engine to produce deductively valid ideas. This has led to several persistent limitations which we describe as reasoning fragility, parsing instability, poor numerology and external tool dependence. Examining the nature of formal reasoning, we explain why such failures are predictable and enduring. We then argue that in order to avoid them, one must develop internalist guarantees of LLM algorithmic performance, and we consider several general forms that such guarantees may take, including direct learning, manual coding, and outsourcing.
Publication Note
This seminar is based on Prof Qingpeng ZHANG’s recent publication in PNAS, with Dr Fei JING—his Postdoctoral Fellow at IDS—as first author. Prof ZHANG will give a detailed account of the paper and its theoretical framework for understanding the predictability limits of complex systems.
Speaker
Prof Boris BABIC
Member, HKU Musketeers Foundation Institute of Data Science
Associate Professor, Department of Philosophy, HKU
Associate Professor (by courtesy), Faculty of Law, HKU
Prof Boris Babic is a Member of the HKU Musketeers Foundation Institute of Data Science, and an Associate Professor at the Department of Philosophy and (by courtesy) the Faculty of Law, HKU. He also serves as an occasional visiting professor in the Decision Sciences department at INSEAD.
Prof Babic received a JD, cum laude, from Harvard Law School, an MS in Statistics and a PhD in Philosophy, from the University of Michigan, Ann Arbor. He also practiced law at Quinn Emanuel Urquhart & Sullivan, LLP in Los Angeles, USA. He completed his postdoctoral fellowship at the California Institute of Technology (Caltech).
Prof Babic’s primary research interests are in Bayesian inference and decision-making, ethics, law, and policy of artificial intelligence and machine learning, especially in medical applications. His research has been published extensively in leading journals such as Science, Nature Machine Intelligence, Nature Digital Medicine, and the Harvard Business Review.
Prof Babic has been on leave from the University of Toronto, where he is Assistant Professor in the Department of Statistics and the Department of Philosophy and a faculty fellow of the Schwartz Reisman Institute for Technology & Society.
For full biography of Prof BABIC, please refer to: https://datascience.hku.hk/people/boris-babic/
Moderator
Prof Qingpeng ZHANG
Member, HKU Musketeers Foundation Institute of Data Science
Associate Professor, Department of Pharmacology and Pharmacy, HKU
Prof Qingpeng Zhang is a Member of the HKU Musketeers Foundation Institute of Data Science and an Associate Professor at the Department of Pharmacology and Pharmacy, HKU. He received the B.S. degree in Automation from Huazhong University of Science and Technology in 2009, and the M.S. and the Ph.D. degrees in Systems and Industrial Engineering (minor in Management Information Systems) from the University of Arizona, in 2011 and 2012, respectively. Prior to joining HKU in 08/2023, he was an Associate Professor with the School of Data Science at The City University of Hong Kong (CityU). He previously worked as a Postdoctoral Research Associate in the Tetherless World Constellation, Department of Computer Science at Rensselaer Polytechnic Institute.
He is a Senior Member of IEEE, a Fellow of the Royal Society of Medicine, and an associate/academic editor for npj Digital Medicine, BMJ Mental Health, INFORMS Journal on Data Science, IEEE TITS, IEEE TCSS, Journal of Alzheimer’s Disease (2021), and PLoS ONE. He was a theme issue editor for Philosophical Transactions of the Royal Society A: Mathematical and Engineering Sciences and the guest editor for a number of other journals. He is on the executive committee of the International Society of Digital Health, Hong Kong Society of Behavior Health, Systems Engineering Society of China, and the Hospital IoT Branch of China Association of Medical Equipment.
His research interests include medical informatics, AI in drug discovery, healthcare data analytics and network science. His research has been published in leading journals such as Nature Human Behaviour, Nature Communications, PNAS, and MIS Quarterly, as well as featured in press such as The Washington Post, The New York Times, BBC, The Times, The Guardian and Ming Pao. He received The President’s Award (2022) and the Outstanding Research Award for Junior Faculty (2021) from CityU and the Andrew P. Sage Best Transactions Paper Award (2021) from IEEE Systems, Man, and Cybernetics Society.
For full biography of Prof Zhang, please refer to: https://datascience.hku.hk/people/qingpeng-zhang/
