?
EXPERIMENTAL GENERATION OF EDUCATIONAL TASKS IN NATURAL SCIENCE DISCIPLINES USING ARTIFICIAL INTELLIGENCE
This study investigates the suitability of modern generative models for the automatic generation of educational task texts. In the first part of the study, we conducted a bibliometric mapping of the research field related to automatic question generation, utilizing three databases: Lens, Dimensions, and the ACM Digital Library. In the second part, we compared the capabilities of three generative systems (ChatGPT-3.5, YaGPT, GigaChat) to formulate various types of assignments based on a textbook content: multiple-choice questions, open-ended questions, and essay topics based on a given text fragment. The source material was a fragment of a fifth-grade biology textbook describing the difference between living and non-living things. The evaluation encompassed an assessment of the models' ability to generate diverse question variants, their proficiency in recording these questions in JSON format for integration into digital platforms, and the correctness of the questions in terms of grammar, relevance, and pedagogical appropriateness.