PestGPT: Leveraging large language models and IoT for timely and customized recommendation generation in sustainable pest management

In response to the escalating food crisis caused by global population growth, precision or smart agriculture has been proposed as a promising solution to achieve efficient and sustainable agricultural production through the Internet of Things (IoT), Data Analytics, and Artificial Intelligence (AI). Although much work has been proposed and proven to be effective in practical application scenarios, such as optimised water utilisation through smart irrigation systems, numerous studies exploring agricultural pest management solutions are limited to pest monitoring and lack the integration of management recommendation generation due to the reliance on agricultural expert knowledge. With the advancement of computer technology, especially for the large language models (LLMs), the combination of IoT and LLMs provides a feasible solution to this limitation by automating the observation data analysis process of agricultural experts. Therefore, this article proposes a framework that integrates with IoT and LLMs to provide users with accurate and customised pest management recommendations based on environmental information from edge devices. In the proposed framework, LLMs incorporate an explicit expert knowledge base to mitigate hallucinations. The proposed framework is deployed in an online farm management platform and is evaluated in a real-world application scenario. After quantitative analysis and validation with real-world application scenarios, we report on the limitations and future directions of the LLMs in the agricultural sector.