DIKU Bits: Cultural Adaptation of Large Language Models

Speaker

Yong Cao, Natural Language Processing (NLP) section

Title

Cultural Adaptation of Large Language Models.

Abstract

“For an average Chinese, do they agree it is a duty towards society to have children? How about for Americans?” Building upon the considerable advances in Large Language Models (LLMs), we are now equipped to address more sophisticated tasks demanding a nuanced understanding of cross-cultural contexts. In my talk, I will discuss how LLMs respond to cultural questions, like the ones posed above, and analyze LLMs’ cultural alignment among diverse human societies. Furthermore, we will delve into two specific domains to examine how contemporary Natural Language Processing (NLP) techniques contribute to enhancing the cultural compatibility of LLMs, including areas of cooking recipes and open-domain dialogues.

Which courses do you teach? (BSc and MSc)

Unfortunately, I am not involved in teaching at the moment.

Which technology/research/projects/startup are you excited to see the evolution of?

I am interested in exploring new techniques aimed at enhancing diversity and cultural awareness in language technologies. Additionally, I am enthusiastic about investigating how LLMs can adapt to different cultures through the incorporation of external world knowledge and user backgrounds.