Technology for creating tests for a math lesson using generative neural networks

the article discusses the issues of using neural networks to generate test tasks for a mathematics course in high school. To generate content for math lessons, it is proposed to use the generative neural network MathGPT, built on the basis of ChatGPT, but additionally trained on a large array of mathematical content. This artificial intelligence system is capable of generating a variety of mathematical content and uses LaTeX syntax to correctly display formulas. The article discusses the issues of generating tasks in a test form on the topic "Actions with ordinary fractions" of a school mathematics course. The issues of generating tests in GIFT format for their placement in the popular free MOODLE software are considered, as well as the possibility of generating a platform-independent test in JavaScript embedded in a page marked up using HTML. The tests developed by artificial intelligence systems were evaluated by mathematics teachers through a focus group. Practical teachers noted the simplicity of preparing tests with a generative neural network in these formats, even for non-specialists in the field of IT. The practical value of this work may be the possibility of preparing individualized control tasks and tasks for self-preparation with the possibility of automated verification and without significant teacher workload.

generative neural networks, GPT, tests, methods of teaching mathematics, educational content, artificial intelligence, task generation