GPT-4 as an Agronomist Assistant? Answering Agriculture Exams Using Large Language Models

This is an academic paper (arXiv preprint, 910 days old, no GitHub repo, zero stars/forks) presenting a benchmarking study rather than a productized tool or reusable component. The core contribution is applying existing LLMs (GPT-4, Llama 2) to a new domain (agriculture certification exams) using standard evaluation methodology. This is incremental: domain-specific benchmarking is a well-established practice; the novelty is limited to the agriculture domain choice. The work has no adoption signals, no software artifact, and provides only reference code accompanying academic research. Platform domination risk is high because OpenAI, Anthropic, and Meta already operate the LLMs being tested, and domain-specific evaluation benchmarks are increasingly commoditized (HELM, OpenCompass, etc.). The paper itself is not a defensible asset—it's a one-time evaluation. There is no market consolidation risk because no commercial product or startup ecosystem exists around this specific work. The displacement horizon is immediate: any competent team can replicate this study in weeks using published LLM APIs and public exam datasets, and platforms themselves conduct such evaluations internally as standard practice. This is a useful academic contribution but carries no defensibility as an open-source project.