Blog CRC1646

The Summer School on Large Language Models for Digital Humanities Research took place from 8^th to 11^th September at the University of Cologne. The meeting was funded by the University of Cologne’s Competence Area III (CA3: Quantitative Modeling of Complex Systems) and organized in cooperation with four institutions: the Center for Data and Simulation Science (CDS), the Department for Digital Humanities (IDH), the Cologne Center for eHumanities (CCeH), and the Data Center for the Humanities (DCH).

The Summer School began with a two-day workshop on prompt engineering and prompting techniques. As this is a rapidly evolving field, staying up to date is especially important. The very engaged and motivating lecturer Christopher Pollin (Digital Humanities Craft OG) presented his latest insights in prompting Large Language Models (LLMs). He shared valuable knowledge on working with LLMs that A05 can apply almost daily in their research.

On the following two days, four different workshop tracks were offered. Emilie Sitter from project A05 attended the track on Using LLMs in Psycholinguistic Research. The lecturers of this workshop, Hanna Woloszyn, Job Schepens and Benjamin Gagl from University of Cologne, introduced strategies on how to apply LLMs in psycholinguistic experiments, which can quickly can become time- and cost-intensive when conducted in-lab with human participants. While offering ideas on how LLMs could be used in psycholinguistic research, the sessions highlighted key differences between human cognition and language models from a cognitive science and psycholinguistics perspective. The main take-away of this workshop was that attempts to augment or replace human experiment data with LLM responses or LLM-derived metrics must be approached with very much caution.

In line with this, the closing keynote by Elen Le Foll (University of Cologne) focused on the downsides of LLMs, including their environmental impact, societal implications such as the use of stolen data and the questionable values they may convey, and potential risks they pose to us as a research community. It provided an important counterbalance to the most of the Summer School, which focused rather on concrete applications and the benefits of LLMs than on the problems and risks they entail. It helped raise awareness of LLMs’ broader impact on research and society.

© Emilie Sitter

© Sascha Hermannski, SFB 1646