## Introduction Large language models are a class of neural models that learn to recognize patterns in natural language. Used for generating text, they respond to questions, and can also be used as tutoring tools. ## Results A recent study examined the behavior of large language models compared to expert human tutors in mathematics. The results show that these models have a similar level of pedagogical quality as experts but use different teaching strategies and linguistic approaches. ## Diversity of Teaching Strategies Large language models tend to use a strategy of pressure for accuracy, but lack feedback and rephrasing, key features of high-quality pedagogy. They also produce longer and more linguistically diverse responses. ## Politeness and Agent Large language models tend to use a more polite and agent-like language, but the results show that this is negatively associated with quality pedagogy. ## Conclusion Large language models have a similar level of pedagogical quality as experts but use different teaching strategies and linguistic approaches. This study highlights the importance of analyzing teaching strategies and linguistic characteristics when evaluating tutoring quality.