Should I Trust AI for Farm Decisions?

Along with Jim Ladlee, I recently reviewed eight different large language model (LLM AI) platforms by asking ten questions related to cucumber production. The questions covered topics such as disease management, pesticide selection, nutrient management, and crop production practices, and the responses were compared to recommendations I would make based on research-based information available through Penn State Extension and our production guides. I rated answers in terms of accuracy and usefulness. Earlier this year, we reported on the study at the Mid-Atlantic Fruit and Vegetable Convention in Hershey, PA. Some tools such as Gemini Free and Grok3 performed relatively well, while others produced incomplete, unclear, insufficient, or even potentially harmful results for our local conditions.

Many of us are learning how to use AI better by writing better prompts offering context, roles, actions, format and tone (CRAFT), and asking follow-up questions for clarity. However, as I found with my informal analysis, sometimes the range of answers to our questions is quite broad, and we need an AI tool that is focused on science-based, locally trusted references.

In January 2026, Penn State Extension launched Tilva^TM, a purpose-built artificial intelligence-powered tool that gives research-based answers to questions. Unlike the general AI LLM tools, Tilva draws exclusively from curated Extension publications and government resources relevant to our state or local area. All the current guides you rely on - the Mid-Atlantic Commercial Vegetable Production guide and the Penn State Tree Fruit Production Guide - are included in Tilva. The articles written by your Extension specialist and local educator are included, as well as databases from the Pennsylvania Department of Agriculture, USDA, and other state Extension publications that are locally important. Thanks to Artificial Intelligence, answers are instantly available in both English and Spanish.

One nice feature of Tilva is that you are able to rate the answer – a thumbs up or thumbs down – and write a comment about what you liked about the answer or what you would like to see improved. The development team takes this feedback seriously, and educators are involved, myself included, to look at why questions were given a particular rating. Because of this attention to detail, Tilva will always continue to improve.

I encourage you to test Tilva. Ask it questions that you are familiar with, or information you may have recently learned at a winter meeting. As with all information you find online, be extra careful when asking questions about pesticide recommendations, financial decisions, or health-related advice, whether human or animal. These types of decisions should be made along with consultation of the pesticide label, financial advisor, or licensed veterinarians and certified health professionals. Tilva is designed to route high-risk questions to Extension educators, but human expertise should always guide final decisions.

If you’re interested in learning more about the informal study Jim and Leah conducted on LLMs, Leah may be contacted at lxf339@psu.edu.