ChatGPT -Based Method for Generating Automobile Accident Reports

With the continuous development of artificial intelligence, the application of pre-trained language model has been widely used in various fields, especially in complex text generation tasks. The fine-tuning is based on OpenAI's GPT -3.5-Turbo-ll06 model and aims to improve its performance in the task of generating autonomous driving accident analysis reports. Using specific accident description data and hand-written summary pairs, the model learns through supervised fine-tuning the ability to extract key details from complex accident descriptions and generate concise, accurate reports. The experimental results show that the overall cosine similarity on the training set reaches 0.78, and the Jaccard similarity reaches 0.22, which can accurately perform specific text generation tasks, and preliminarily proves the effectiveness of the fine-tuning model.