China has launched its first open-source large language model (LLM), Green Shield, for better crop protection. 

Developed by Nanjing Agricultural University (NAU) in partnership with the National Key Laboratory of Agricultural Biosafety and over 30 industry institutions, the model is designed to offer scientific agricultural guidance and ensure proper pesticide use, Science and Technology Daily reported on May 26. 

"China faces frequent crop pest outbreaks and the issue of pesticide resistance," said Dong Shameng, project leader and vice dean of NAU's College of Plant Protection. 

He was further quoted by the newspaper as telling attendees at the launch on May 26: "Farmers urgently need professional guidance at the grassroots level. However, general-purpose LLMs often provide inaccurate answers to plant protection questions and, more critically, give poorly standardized, sometimes risky advice on pesticide use." 

To solve the problem, the team built a specialized corpus of over 2.5 billion tokens from academic papers, patents, national standards and field reports. The corpus covers major crops, including rice, wheat, soybeans, vegetables and fruit trees, and integrates information on pest monitoring, green control measures and pesticide registration, according to the report. 

Wang Dongbo, a professor at NAU's College of Information Management, said the model can precisely identify crop types, growth stages and symptoms of crop diseases. It then generates integrated growth control strategies. 

"With targeted training, the model converges well and recognizes pests with high precision," said Wang. 

Before presenting any recommendation, the model automatically cross-references the national pesticide registration database to verify each chemical against banned lists, approved crops and dosage limits. Any non-compliant suggestion is blocked and self-corrected, preventing pesticide misuse at the source, said Wang. 

Wang Yuanchao, vice president of NAU, said the university will continue field tests and model iterations to create an intelligent tool that is "understandable, usable and effective" for farmers, empowering modern agriculture with digital technologies across the entire chain.