OxTalks is Changing
OxTalks will soon be transitioning to Oxford Events (full details are available on the Staff Gateway). A two-week publishing freeze is expected in early Hilary to allow all events to be migrated to the new platform. During this period, you will not be able to submit or edit events on OxTalks. The exact freeze dates will be confirmed as soon as possible.
If you have any questions, please contact halo@digital.ox.ac.uk
Uses of LLMs in Social Sciences: Benefits and Risk
This presentation outlines a constrained use of large language models (LLMs) in sociology and demography, grounded in an information-theoretic account of language. LLMs are treated as lossy statistical compression systems operating over non-injective mappings from social meaning to text.
On this basis, their appropriate uses are limited to supporting existing analytical work: improving clarity of expression, surfacing textual regularities, and stress-testing arguments and assumptions. They are not sources of evidence, explanation, or social inference. Used within these limits, LLMs can reduce linguistic friction without being mistaken for epistemic agents.
Biography:
Daniel is a Senior Data Scientist and Postdoctoral Researcher in Computational Social Science at the Leverhulme Centre for Demographic Science. His research in the Centre is focused on the development of robust estimation methods for social science and in the development of software libraries in Python and R to perform multiverse-type estimations.
Additionally, he researches the application of machine learning / deep learning models (e.g., BERT, RoBERTa, GPT-2) on social science problems like misinformation detection and characterisation on social media text, and the characterisation of social movement emotions over time based on associated tweets.
Prior to joining Oxford, Daniel completed his PhD in Computational Social Science at the University of Leeds, and before that worked for several years as a quantitative analyst at the Pontifical Catholic University of Chile. His general interests are related to the use of machine learning methods to understand human behaviour and the application of novel methods for robust parameter estimation, either using multiverse-type approaches or Bayesian / probabilistic approaches.
Please join either in person or online. For in-person attendees, the talk will be preceded by a light lunch at 12.15pm.
Date:
16 February 2026, 12:45
Venue:
Seminar Room, Department of Sociology and Online
Speaker:
Daniel Valdenegro (University of Oxford)
Organising department:
Department of Sociology
Organiser contact email address:
comms@sociology.ox.ac.uk
Part of:
Sociology Department Weekly Seminar
Booking required?:
Not required
Audience:
Members of the University only
Editor:
Laurel Millen-Quinn
