Oxford LLMs 2024 Highlights
In late September 2024, Nuffield College, supported by the Oxford Van Houten Fund, ran an intensive five-day workshop on AI for social science. Early-career researchers attended lectures on large language models, completed hands-on Python tutorials, and collaborated on research projects with AI experts.
We received a little short of 200 applications and selected 30 participants.
Workshop Themes and Materials
Our lecture series, developed by Grigory Sapunov (Intento), Tatiana Shavrina (Meta), and Ilya Boytsov (Wayfair), covered LLM foundations, transformer architecture, in-context learning, and emergent reasoning. We introduced advanced techniques such as agent-based systems, fine-tuning methods, and self-hosting considerations. Applied tutorials demonstrated retrieval-augmented generation for social science data and observability tools for model monitoring. All lecture slides, notebooks, and code samples are available in the 2024 materials archive on the materials page and GitHub.
Established scholars from political science, sociology, and history - Lisa P. Argyle, Chris Barrie, Raymond Duch, Thomas Hegghammer, Neil Ketchley, Alison Koh, and Alexis Palmer - shared examples of research using LLMs from their work. Guest presenters from companies like Qdrant, Ori Cloud, Arize, and Google led practical seminars, with contributions from Atita Arora, John Githuly, Christian Silva, and Ciera Fowler.
Collaborative Research Projects
All participants contributed to the main project: developing an approach for survey data preprocessing aimed at training LLMs to predict people’s opinions and self-reported behaviors. Ori Cloud provided GPU support for this project. The models and datasets are published on Oxford LLMs Huggingface profile, and the group is polishing the work before submission.
Looking Ahead
Feedback highlighted that the collaborative project was the workshop’s most valuable component and that participants wanted more coding exercises. As a result, the September 2025 workshop will be slightly more technical, with increased focus on coding sessions and research projects.
If you are an early-career researcher eager to master LLM architectures, coding workflows, and participate in interdisciplinary research projects, apply for the 2025 workshop!