Adapting Historical informatics Control Tasks with Large Language Model Potentials: practical experience and methodological findings

The integration of Large Language Models (LLMs) into the pedagogical framework of Historical Informatics marks a definitive shift from reproductive learning toward the design of sophisticated analytical pipelines. This article substantiates a conceptual approach to adapting academic assignments by transforming them into technical specifications (prompts) for LLMs, prioritizing the verifiable processing of primary historical and historiographical sources over mere text generation. In this paradigm, algorithmic thinking and the ability to structure complex queries become essential components of a historian’s competencies in the digital era, reflecting a shift toward new forms of information interaction. The study draws on the practical application of six distinct types of control tasks, including automated bibliographic retrieval, the extraction of linguistic variables to quantify historical indicators, and the construction of multi-level domain-specific thesauruses. The author demonstrates that the methodological value of prompt engineering in the humanities lies in the necessary explication of research procedures—ranging from the verbal description of visual data trends in the Google Ngram Viewer to the implementation of few-shot techniques for standardizing named entity recognition. Furthermore, the paper addresses critical pedagogical challenges, specifically the "hallucination of plausibility" that diminishes student skepticism, and the technical management of false-positive and false-negative model outputs, which necessitates cross-model verification strategies. Ultimately, the research posits an inevitable transformation of the instructor's role into a moderator of the student’s intellectual engagement with AI, where the depth and structural precision of a prompt serve as the primary metric of domain expertise and subject immersion.