Development of a Method for Using Graphical Diagrams for Queries to Large Language Models

ENG: This paper is devoted to the development of an innovative method that combines the visual structure of graphic diagrams with the power of large language models (LLMs). Based on experiments with GPT-4, Claude 3, and Gemini 1.5 Pro, it has been demonstrated that the use of formalized diagrams as queries increases the coherence and accuracy of responses by 16–25% compared to traditional text instructions. The proposed approach opens up new opportunities for structured human-artificial intelligence interaction in education, science, and analytics.