Abstract: Over the past two decades, language modeling has emerged as a primary methodology for both language understanding and generation. It has garnered significant scholarly attention as a pivotal component within the field of Natural Language Processing (NLP). In recent years, the advent of Large Language Models (LLMs), exemplified by technologies like ChatGPT, has marked remarkable advancements and exerted profound influence on the transformation and progression of artificial intelligence and other domains. Given the rapid evolution of LLMs, this paper initiates with a comprehensive review of the developmental trajectory pertaining to LLMs, including technical architecture and model scalability. Subsequently, it consolidates various aspects of model training techniques, optimization methodologies, and evaluation criteria. The paper proceeds to highlight specialized technical tools prevalent within this field, scrutinizing the current application landscape of LLMs across domains such as education, healthcare, finance, and industry. It delves into an analysis of their strengths and limitations. Furthermore, the paper undertakes an in-depth exploration of critical issues precipitated by large language models in the realms of social ethics, privacy, and security, while also proposing viable solutions to address these concerns. In conclusion, the paper provides an outlook on future research directions for large language models, encompassing areas like model scalability and efficiency, multimodal processing, and their societal impact. By conducting a comprehensive assessment of the present research landscape and forthcoming trends, this paper aspires to furnish researchers with profound insights and inspiration in the realm of large language models.