Mark Dingemanse

Radboud University Nijmegen

Homepage

Talk: What makes LLMs so irresistible?

Abstract

Large language models of the instruction-tuned kind have taken the world by storm. I explore how their chat-based design and well-formed responses condition users to over-attribute agency and intelligence and obscure how the burden of interpretation and calibration falls mainly to the human side. I will argue that linguistic analysis in relation to LLMs should include a critical understanding of this technology, and that choices for or against deployment of any technology should be guided by core principles of scholarly conduct, most importantly transparency, honesty, scrupulousness, independence, and responsibility (Netherlands Code of Conduct for Research Integrity). Finally, I will review our work on the European Open Source Generative AI Index, which has spurred the development of transparent and reproducible approaches to research on large language models.

Bio

Mark Dingemanse is Associate Professor in Language and Communication at Radboud University Nijmegen. He is also PI of Futures of Language, a transdisciplinary research programme that looks at artisanal and artificial ways of languaging and that builds on our earlier work in Elementary Particles of Conversation.