Abstract

Large language models (LLMs) have rapidly become a conventional tool across education, business, and public administration. Their implications for social and sciences and humanities are equally profound, yet their integration into everyday research practices remains uneven and often poorly understood.

This workshop introduces participants to the use of LLMs as a methodological tool in both quantitative and qualitative social research. Moving beyond abstract discussions of artificial intelligence, the workshop focuses on practical workflows that allow researchers to incorporate LLMs into data collection, processing, and analysis.

The workshop is particularly relevant for research in Central Asia, where limited data availability often constrains empirical work. LLMs open new possibilities by enabling researchers to work with unstructured text, generate structured datasets, assist in coding qualitative materials, and augment existing data sources.

Participants will be introduced to the R package ellmer, which provides a flexible interface for working with LLMs in reproducible research environments. Through hands-on examples, we will demonstrate how LLMs can be used for tasks such as information extraction, text classification, synthetic data generation, and exploratory analysis.

During the workshop we also discuss the limitations of LLMs, including biases, reproducibility challenges, and their uneven performance across languages relevant to the Central Eurasian region.

Participants are expected to have RStudio installed and possess basic familiarity with R or Python.