Meet the Experts: LLMs - Opportunities and Challenges for the Social Sciences

Large Language Models (LLMs) have revolutionized natural language processing and offer exciting opportunities for the social sciences. This talk explores the potential of LLMs in enhancing research methods for data analysis and enrichment that are ultimately used to study social science research questions and in GESIS services. In particular, we look at how LLMs-based techniques such as zero-shot and few-shot learning can be applied to produce training data for methods for opinion mining, survey item tagging, and variable extraction. Alongside the opportunities, we also discuss the challenges and risks associated with the utilization of LLMs, including robustness, model decay, hallucination and privacy, and ethical issues. Finally, in this presentation, we also discuss the possible applications of LLMs for established GESIS services such as GESIS search but also for services under development, e.g., novel data offers created from Web content and social media or the GESIS Methods Hub, a new platform for sharing computational methods and models for analysis of digital behavioral data.
Presenters:
Dr. Dimitar Dimitrov
Dr. Hajira Jabeen

[@MBison-oc5tc]

how does finetuning work?