Large Language Models (LLMs) are transforming how we can approach research in the social sciences and humanities. But alongside their generalization power come serious methodological considerations. In this full-day workshop we will first learn more from three speakers in the morning session about the capabilities and pitfalls of LLMs, how they work and how they can be evaluated. Then, in the afternoon session, we will use named entity recognition (NER) as a use-case to see how LLMs might assist us in improving a baseline NER model by generating additional, synthetic training material. We will make use of historical data. While the use-case itself only serves as an example, the methodology used (expanding training data with LLMs, evaluating and comparing a base model to an improved model) can be applied broadly to different topics and fields.

Target audience

This workshop is explicitly targeted towards researchers in the humanities and social sciences. The first session does not require any prior knowledge. The second session in the afternoon is hands-on and technical, where we will run through a prepared notebook on Google Colab together (so you will need an account on https://colab.google/). Experience with Python and Jupyter notebooks is recommended for this second part.

Registration

You have to register for the separate sessions described below (scroll down on these links and click “Aanmelden”). You do not have to attend both!

• Morning session 09:00-12:00 (theoretical): https://ivdnt.org/evenement/workshop-llms-1/
• Afternoon session 13:00-17:00 (practical): https://ivdnt.org/evenement/workshop-llms-2/

More information

Learn more on the instituut voor de Nederlandse taal website