How ChatGPT Is Shaping Next-Generation Patent Solutions

Abstract

With the rapid advancement of artificial intelligence, large language models (LLMs) such as ChatGPT have garnered increasing attention in the field of text generation. However, when applied to patent drafting in the architectural domain, these models often produce verbose texts and omit critical elements, leading to issues such as limited claim scope and insufficient disclosure in the specifications. To address these challenges, this study conducted experimental research on the application of AI technologies, including GPT-4o, in the drafting and publication of patents related to prefabricated integral buildings. The results indicate that GPT-4o performs well in generating claims and technical descriptions. Nevertheless, problems such as repetitive descriptions of technical achievements, ambiguity in result interpretation, and a lack of detailed implementation methods were observed during the generation process. Therefore, we recommend incorporating expert review mechanisms when using GPT-4o for patent drafting, to enhance the accuracy and professionalism of the output. Moreover, enriching training datasets and enforcing a rigorous textual review workflow can further optimize model outputs, aligning them more closely with the high standards required in patent documentation.

Ähnliche Arbeiten