This paper provides a comprehensive overview of the integration of Large Language Models (LLMs) into fuzzing test techniques, a traditional method for software testing. Fuzzing, which involves generating random inputs to test software reliability and security, has evolved significantly over the years. The rapid development of LLMs has opened new possibilities for enhancing fuzzing tests, making them more efficient and accurate. The paper discusses two main research topics: "Fuzzer by LLM" and "Fine-Tuning Fuzzer." "Fuzzer by LLM" focuses on integrating LLMs into the seed generation and mutation processes to improve the performance of traditional fuzzing tests. "Fine-Tuning Fuzzer" involves using LLMs as the core fuzzer, leveraging their learning capabilities to generate abnormal inputs and effectively test deep learning libraries. The paper also compares LLM-based fuzzers with traditional fuzzers, highlighting advantages such as higher API and code coverage, computational efficiency, and the ability to detect more complex errors. It discusses the potential for increased automation in fuzzing tests and the challenges, including hallucinations, computational efficiency, and the need for well-defined evaluation metrics. Finally, the paper explores future directions, such as the potential for LLM-based fuzzing in hardware testing and the development of a universal evaluation framework for LLM-based fuzzing tests. The insights and findings are intended to guide researchers and practitioners in advancing the field of automated software testing using LLMs.This paper provides a comprehensive overview of the integration of Large Language Models (LLMs) into fuzzing test techniques, a traditional method for software testing. Fuzzing, which involves generating random inputs to test software reliability and security, has evolved significantly over the years. The rapid development of LLMs has opened new possibilities for enhancing fuzzing tests, making them more efficient and accurate. The paper discusses two main research topics: "Fuzzer by LLM" and "Fine-Tuning Fuzzer." "Fuzzer by LLM" focuses on integrating LLMs into the seed generation and mutation processes to improve the performance of traditional fuzzing tests. "Fine-Tuning Fuzzer" involves using LLMs as the core fuzzer, leveraging their learning capabilities to generate abnormal inputs and effectively test deep learning libraries. The paper also compares LLM-based fuzzers with traditional fuzzers, highlighting advantages such as higher API and code coverage, computational efficiency, and the ability to detect more complex errors. It discusses the potential for increased automation in fuzzing tests and the challenges, including hallucinations, computational efficiency, and the need for well-defined evaluation metrics. Finally, the paper explores future directions, such as the potential for LLM-based fuzzing in hardware testing and the development of a universal evaluation framework for LLM-based fuzzing tests. The insights and findings are intended to guide researchers and practitioners in advancing the field of automated software testing using LLMs.