Using Advanced LLMs to Enhance Smaller LLMs: An Interpretable Knowledge Distillation Approach
21 Feb 2025
12:30 -14:30
Times are shown in local time.
Open to: All
Room W2.02 (Cambridge Judge Business School)
Trumpington St
Cambridge
CB2 1AG
United Kingdom
Join our Marketing seminar
Speaker: Professor K Sudhir, Yale University
About the seminar topic
Advanced Large language models (LLMs) like GPT-4 or LlaMa 3 provide superior performance in complex human-like interactions. But they are costly, or too large for edge devices such as smartphones and harder to self-host, leading to security and privacy concerns. This paper introduces a novel interpretable knowledge distillation approach to enhance the performance of smaller, more economical LLMs that firms can self-host. We study this problem in the context of building a customer service agent aimed at achieving high customer satisfaction through goal-oriented dialogues. Unlike traditional knowledge distillation, where the ‘student’ model learns directly from the ‘teacher’ model’s responses via fine-tuning, our interpretable ‘strategy’ teaching approach involves the teacher providing strategies to improve the student’s performance in various scenarios.
This method alternates between a ‘scenario generation’ step and a ‘strategies for improvement’ step, creating a customised library of scenarios and optimised strategies for automated prompting. The method requires only black-box access to both student and teacher models. Hence it can be used without manipulating model parameters. In our customer service application, the method improves performance, and the learned strategies are transferable to other LLMs and scenarios beyond the training set. The method’s interpretability helps safeguard against potential harms through human audit.
Speaker bio
Professor K Sudhir is James L Frank Professor of Private Enterprise, Management and Marketing and Director of the Yale China India Insights Program (CIIP). His research focuses on gaining market insights by analysing consumer and firm actions through econometric modelling. As director of the China India Insights Program, he specialises in research on consumers in emerging markets. He has consulted for Fortune 500 US firms and Indian firms across many industries such as technology, financial services, entertainment and retailing, specialising in analysing their internal data to obtain actionable market insights. He leads the various data-driven research projects at the Yale Center for Customer Insights.
Professor Sudhir’s research has been honoured with numerous awards across all major quantitative marketing journals. Two of his papers published in 2001 were among the top 10 finalists for papers with the most Long Term Impact published in Marketing Science and Management Science in the previous 10 years from 2009 to 2011. He has received the Bass and Little Awards at Marketing Science and the Lehmann Award at the Journal of Marketing Research and honourable mentions for the Wittink Award in Quantitative Marketing and Economics and Best Paper Award in International Journal of Research in Marketing.
He has also been a finalist for the Paul Green Award at the Journal of Marketing Research. Professor Sudhir is currently the Editor-in-Chief of Marketing Science. He previously served as Senior Editor at Marketing Science and Associate Editor at all of quantitative marketing’s leading journals, the Journal of Marketing Research, Management Science, Marketing Science, and Quantitative Marketing and Economics.
Register
No registration required. If you have any questions about this seminar, please email Luke Slater.