Large Language Models for Health and Social Science Research

Date: 29th June through 3rd July.

By: Daniel Valdenegro (Teaching Assistant to be Confirmed)

Location: Nuffield College, University of Oxford

Course Description: Artificial Intelligence (AI) and Large Language Models (LLMs) have become ubiquitous concepts in research and everyday life. The demand for a clear, grounded understanding of the capacities and shortcomings of these tools will only grow as their popularity increases. In this 5-day intensive course, we will explore the fundamental theoretical and practical aspects of using Large Language Models in health and social science research, maintaining a grounded and critical stance. We will examine the origins and development of the core architectures powering today’s most widely used language models, as well as their current applications in research, complemented by practical sessions on how to work with both commercial and locally hosted models.

Each day runs from 9:30 to 16:30, with:

Morning lectures (9:30–12:30): Concepts, theory, and methods.
Lunch (12:30–13:30): Provided at Nuffield College.
Afternoon practicals (13:30–16:30): Guided coding sessions, data activities, research talks, and group exercises By the end of this course, you will gain:

1. A working knowledge of how LLMs are trained, evaluated, and applied in research.
2. Practical skills in accessing, fine-tuning, and applying LLMs to real-world datasets.
3. Critical tools to assess bias, ethics, and reproducibility in LLM-driven research.
4. Exposure to interdisciplinary projects at the intersection of AI, health, and the social sciences.

Lunch and refreshments are provided daily, fostering an informal, collaborative learning environment in the heart of Oxford’s academic community. Course content by day:

Day 1: Foundations – Introduction to NLP and initial language models: tokenization, embeddings, RNN-LSTM vs Attention, Text-generation Transformer Models.
Day 2: Applications – Advent of Large, transformer-based, text-generation language models: Emerging capacities in summarization, classification and information extraction. Review of application in health and social science research. Practical session on this tasks.
Day 3: Use: Review of current commercial and open-source models. Review of the API use and local hosting options.
Day 4: Ethical considerations: Review of current research on technical limitation of LLMs. Review of current research con ethical challenges of on the Use of LLMs. Practical session exploring those limitations.
Day 5: Future Directions – Research frontiers: multimodal models, causal inference, and integrating LLMs into scientific workflows.

Attendance will be recognised through Accredible badges.

Information on how to register for this course and for information on course fees will be published in due course.

For any queries, contact teaching@demographic.ox.ac.uk.