Proceedings of the International scientific and practical conference “Science, Technology and Culture: Strategies for Sustainable Development” (December 15-17, 2025) / Publisher website: www.naukainfo.com. – Krakow, Poland, 2025. – 120 p.

74 Large Language Models (LLMs) are deep neural networks trained on vast corpora of textual data. They function as statistical prediction engines capable of generating language with rich contextual awareness and elements of reasoning. A critical catalyst in the development of LLMs is the Transformer architecture. Through the self-attention mechanism, this architecture has enabled efficient parallelization of training on unprecedented data scales. It has also allowed LLMs to process longer contextual sequences, which is essential for maintaining coherence in extended dialogues. To ensure that LLM outputs meet ethical and pedagogical expectations and maintain accuracy, Reinforcement Learning from Human Feedback (RLHF) is employed, optimizing model behavior through human-in-the-loop evaluation. The integration of LLMs has enabled deep personalization and the creation of authentic communicative scenarios, addressing key limitations of earlier CALL systems. Trained on billions of words, LLMs are capable of generating context-aware interactions that simulate spontaneity, thereby bringing online language practice closer than ever to real-world linguistic immersion. LLMs can automatically adapt tasks, explanations, and communicative scenarios to a learner’s current level of proficiency. Moreover, the integration of LLMs has led to a transformative breakthrough in the development of productive language competencies, namely speaking and writing. This breakthrough lies in the shift from discriminative systems, which merely evaluated correctness, to generative systems capable of producing adaptive and authentic communicative scenarios in real time. LLMs have opened new horizons for speaking practice, which has traditionally required the continuous presence of a human tutor. Specialized applications (e.g., Speak ) leverage advanced AI technologies by functioning as round-the-clock conversational partners, enabling users to practice on any topic while simulating the spontaneity of natural communication. AI tutors provide immediate and detailed feedback through speech analysis. This feedback encompasses the assessment of pronunciation and intonation (including