A Survey of Large Language Models

This paper is a systematic survey of large language models, motivated by their rapid evolution and the need for new frameworks to understand their development and impact. It organizes recent technical advances along four dimensions: pre-training methodologies that establish core capabilities through large-scale self-supervised training, architectural innovation, and data curation; post-training techniques including supervised fine-tuning and reinforcement learning for task adaptation, alignment, and safety; utilization strategies such as in-context learning, prompt engineering, and agentic reasoning; and evaluation methods spanning benchmarks for language, reasoning, and safety.

Beyond cataloguing methods, the survey identifies critical research issues, including theoretical foundations, efficient scaling, alignment, and agentic capability, and highlights the open challenges they pose. By synthesizing state-of-the-art insights and emerging trends, it aims to provide a comprehensive, structured framework for understanding the trajectory, current limitations, and future directions of LLM progress. This mattered as a reference point consolidating a fast-moving field for both researchers and practitioners.