Imagine an AI that could predict the effectiveness of food policies before they're even implemented. Researchers are exploring just that, using Large Language Models (LLMs) to forecast the impact of interventions aimed at changing our eating habits. By feeding the LLM details about past studies—like location, target group, and type of intervention—they've built a system called PREDICT that can guess the direction of an intervention's outcome with almost 80% accuracy. This means it can predict whether a policy will increase or decrease things like food waste or meat consumption. The magic happens through a process called fine-tuning, where the LLM learns from a massive dataset of over 200 past interventions, totaling over 2 million observations. The research reveals that the way you talk to the AI matters. Simple, direct prompts work best, while overly detailed instructions can confuse it. There's also a sweet spot for the amount of training data: too much or too little can decrease accuracy. While PREDICT shows promise, there are challenges. It struggles with entirely new scenarios it hasn't seen before, emphasizing the need for diverse training data. The real test will be using it in real-world policy decisions. Imagine using AI to tackle food waste, promote healthier eating, or even address climate change through food policy. This research opens the door to using AI not just to understand our past food choices, but to shape our future ones.