Research on the Technological Evolution Path and Realistic Dilemma of Large Language Models

Abstract

With the rapid advancement of deep learning, artificial intelligence is shifting from traditional rule- and statistics-based methods to frameworks centered on deep learning. Large-scale pre-trained language models, with their broad applicability, have achieved notable gains in education, healthcare, office work, and other fields. However, at present, these models still face challenges such as illusion generation, lagging knowledge update and ethical risks. This paper aims to discuss the development and application of the large language model based on Transformer architecture, and analyze its performance in various fields and the main problems it faces. Through combing and analyzing the relevant literature, this paper mainly explores the model development, pre-training and fine-tuning strategies based on Transformer architecture, as well as application examples. The results indicate that the large language model has obvious advantages in generating and understanding, but there are still obvious shortcomings in dealing with ethics, safety and timeliness. Future research should emphasize multi-modal integration, value alignment, and efficiency enhancement, ensuring that the technology can be used safely and reliably across various fields, while maximizing social benefits via the effective interplay of technology, ethics, and law.