A Domain-Targeted Question-Answering Dataset for Food and Nutrition Applications
Evaluation of four general-purpose LLMs on FoodBench tasks shows limited performance across nutrient estimation, traffic-light classification, and food interoperability, even with few-shot prompting, highlighting the need for domain-specialized LLMs fine-tuned on food data.