During this hybrid lecture, professor Antal van den Bosch will discuss the utilization of the latest open-source large language models (LLMs).

Over time, both closed and open-source LLMs, the technological foundation of chatbots like ChatGPT, have become increasingly accessible for research purposes, thanks in part to platforms like Huggingface.

In this lecture, Van den Bosch will explore various applications of open LLMs. For example, how a pretrained LLM for Dutch, RobBERT, can be tested on existing cloze test data, as it is trained on masked word prediction; and how BART, a transformer for text-to-text processing, can be fine-tuned to become a Dutch spelling corrector. Additionally, Van den Bosch will address one of the key hurdles in being able to flexibly work with LLMs: The sparse availability of GPUs and the energy-consuming deep learners.

This lecture offers a hybrid format, allowing you to attend either in person at the Digital Humanities Workspace (0.32) in the University Library City Centre or online via Teams.

About the speaker

Antal van den Bosch is a professor of Language, Communication, and Computation at the Faculty of Humanities at Utrecht University. His research focuses on text mining, machine learning, and computational linguistics. Van den Bosch serves as a strategic advisor for the Centre for Digital Humanities. He also chairs the Domain Social Sciences and Humanities and is a member of the Dutch Research Council (NWO) Executive Board.