Entrepreneurship education is critical in encouraging students' innovation, creativity, and entrepreneurial spirit. It provides essential skills and knowledge, enabling them to open their creative potential and apply innovative thinking across diverse professional fields. With the widespread application of large language models in education, intelligent-assisted teaching in entrepreneurship education is stepping into a new learning phase anytime and anywhere. Entrepreneurship education extends across interdisciplinary knowledge fields, incorporating subjects like finance and risk management, which require advanced mathematical computational skills. This complexity presents new challenges for artificial-intelligence-assisted question-and-answer models. The study explores how students can maximize the knowledge repository of current large language models to improve learning efficiency and experimentally validates the performance differences between large language models and graph convolutional reasoning models regarding the complex semantic reasoning and mathematical computational demands in entrepreneurship education questions. Based on case studies, it is found that despite the broad prospects of large language models in entrepreneurship education, they still need to improve in practical applications. Especially in tasks within entrepreneurship education that demand precision, such as mathematical computations and risk assessment, the accuracy and efficiency of existing models still need improvement. Therefore, further exploration into algorithm optimization, model fusion, and other technical enhancements can improve the processing capabilities of intelligent question-and-answer systems for specific domain issues, aiming to meet the practical needs of entrepreneurship education.