METHODOLOGICAL FOUNDATIONS OF INTEGRATING GENERATIVE LANGUAGE MODELS INTO TEACHING PROGRAMMING FOR STUDENTS OF TECHNICAL SPECIALTIES

Abstract

The article substantiates a differentiated methodological framework for integrating generative language models into programming courses for students of engineering, physical-mathematical, energy, and applied-mathematical specialties. The basic concepts of the subject field are delineated (generative language models, AI programming assistants, intelligent tutoring systems, adaptive platforms), and a scale of integration is formulated, ranging from episodic use to systemic course redesign. The transformation of programming education goals under the availability of large language models is analyzed: the emphasis shifts from syntax and template algorithms to task decomposition, requirements specification, interpretation of generated code, its verification and debugging. Four didactic integration models are developed: pair programming with the model as driver, LLM as opponent in structured debates, flipped classroom with AI-generated examples, structured prompt-tasks with a query quality rubric. The necessity of restructuring the assessment system through process portfolios, oral defense of code, metacognitive tasks, and differentiated AI access modes in different course modules is justified. The specificity of technical-specialty students as recipients of integration is identified: the risk of premature automation before forming algorithmic intuition, and the symmetric risk of LLM underestimation in specialties with instrumental use of code. The institutional conditions of integration in Ukrainian higher technical education are considered in the context of post-war reconstruction and European integration, compared with framework documents of the European Commission, UNESCO, and OECD. A three-level typology of integration modes (initial, intermediate, and advanced) is proposed, differentiated by the role of programming in the specialty and the stage of the educational cycle.