Synthetic data
Can LLMs generate respondent panels that pass standard quality checks? When they do not, where do they fail, and can we measure it before we use them?
We run an annual summer school at Nuffield College, develop an open-source course on the fundamentals of LLMs for social scientists, and work on methods research. Our faculty spans academic research and the ML industry. Everything is open.
We develop methods for applying language models to social science research. Code and datasets are open source on GitHub and Hugging Face.
Can LLMs generate respondent panels that pass standard quality checks? When they do not, where do they fail, and can we measure it before we use them?
How well do open-weights models recover real survey responses across languages, age cohorts, and political contexts, and what drives the gaps?
Building eval sets that survive contamination, treating annotator disagreement as a measurement signal rather than noise.
Early-career researchers. Lectures, labs, collaborative research. Now entering its fourth year.
Lectures on LLM fundamentals, evaluation, fine-tuning, and AI safety. A Kaggle-style competition ran alongside a collaborative research project.
A five-module course mirroring the summer-school syllabus. Taught by invitation at Oxford, ESSCA Paris, and the EUI. All materials on GitHub.
Oxford · DPIR
ESSCA · Paris
EUI · Florence