Easily accessible (small) LLMs in historical studies: opportunities, limitations, pitfalls

Automation of certain parts of data collection and data processing in the field of Social Sciences and Humanities would enable researchers to skip some of the more painstaking tasks, leaving more time for the actual analysis and opening up the possibility to work with larger data sets. With the current and imminent generations of open-source Large Language Models (LLMs) and small Large Language Models (sLLMs) it seems already attainable for individual researchers to speed up onerous but necessary tasks such as qualitative data coding using personal computers, while keeping control of their datasets at all times. Using the qualitative coding process of the former Common Rules project of research group Social Enterprises & Institutions for Collective Action (SEICA, Erasmus University Rotterdam) as a case study, we explored whether open source, low-threshold (s)LLMs are already able to perform specific tasks such as qualitative data coding, and what are the limitations and pitfalls that current and future researchers in the field of Social Sciences and Humanities have to reckon with when applying such AI-driven aides.