Event overview
We are delighted to welcome Dr Jamie Cummins, University of Bern,to give a talk entitled: How (not) to use large language models in social science research.
Abstract: Generative large language models are seeing an explosion of use in the social sciences, driven by their accessibility, easy of use, and capacity for application to an unprecedented range of tasks. However, the accessibility of LLMs obscures the fact that there are a range of important decisions that must be made when using them, and these decisions can dramatically affect output. Further, many of these decisions are gated off from users in commercial chat interfaces, and other features of these interfaces (e.g., the chat context) lead to erroneous mental models of how LLMs “work”. In this talk, I will describe some of these issues, as well as other pitfalls related to the use and interpretation of language models applied to social science research questions. In particular, I will describe examples from two use cases of language models: to model cognitive dissonance effects, and to act as stand-ins for human participants. I will describe how we can more rigorously and systematically apply and evaluate LLMs in research.
Bio: I am a meta-scientist working at the University of Bern, Switzerland, currently a visiting scholar at University of Oxford, UK. Before this, I spent several years working as a postdoc and PhD student in psychology at Ghent University, Belgium. I am the developer of RegCheck (https://regcheck.app) and other research trustworthiness assessment tools, and am more generally interested in forensic metascience and the use of large language models in research.
Please join us in person if you can. Or click here to join the lecture on Teams.
Dates & times
| Date | Time | Add to calendar |
|---|---|---|
| 22 Jan 2026 | 4:00pm - 5:00pm |
Accessibility
If you are attending an event and need the College to help with any mobility requirements you may have, please contact the event organiser in advance to ensure we can accommodate your needs.
