Research on the Problem-Solving Ability of Large Language Models in Mathematics Questions

In recent years, an increasing number of large language models have emerged and been extensively applied in mathematical problem - solving. Their remarkable computational capabilities not only enable the rapid calculation of complex equations but also provide in - depth logical analysis, offering substantial assistance to humans dealing with intricate mathematical challenges. However, the vast diversity of mathematical domains, ranging from algebraic number theory to differential geometry, demands highly specialized training frameworks and unique datasets tailored to specific problem types. This poses multifaceted challenges, such as data sparsity in niche areas and the need for domain - specific knowledge integration. This study will primarily investigate three key areas: a comprehensive overview of large language models, including their underlying architectures; their application status in various mathematical domains, like how they are used in financial mathematics for risk assessment; and the factors influencing the quality of mathematical reasoning approaches, such as the impact of training data size. Additionally, this research will summarize existing superior training methods, such as fine - tuning techniques, and provide directions for the future development of large language models in mathematical problem - solving, for example, exploring more efficient neural network architectures.