Specialized language model for personalized training of future engineers in the academic discipline "Mathematical Analysis"

the article discusses the types of large language models used for teaching mathematical analysis to future engineers. An assessment was made of how the answers from three large language models align with the characteristics of educational content in the context of a basic course in mathematical analysis. Typical errors made by the models are highlighted by comparing their answers with methodological materials developed at the Scientific and Educational Center for Mathematics of ITMO University. The necessity of creating a specialized language model is established. Methods for creating such a model are shown, along with its advantages and disadvantages. The most suitable approaches to creating a specialized language model within the education system are identified: the development of a new large language model, the use of platform solutions, and the fine-tuning of an open-source large language model. It is shown that creating a new large language model and using paid platform solutions requires significant resources. Fine-tuning models can be carried out at universities with the efforts of individual lecturers or interested research groups. A comparative analysis of open-source models is conducted. Possible fine-tuning algorithms and programming languages for five modern open-source large language models are identified.

personalized learning, future engineers, mathematical analysis, artificial intelligence, large language model, hallucination, specialized language model, fine-tuning