Stochastic Streets: A Walk Through Random LLM Address Generation in four European Cities

Stochastic Streets: A Walk Through Random LLM Address Generation in four European Cities This repository contains the results generated by six Large Language Models (LLMs): GPT-4.1, Gemini 2.5 Flash, Gemma 3-12B, LLaMA 3.1–8B, Mistral-7B, Qwen3-8B,using prompts to generate random complete addresses in four different cities: Madrid, Rome, Paris, and Amsterdam. Each model generated 1000 addresses per city under two conditions: City-specific prompt in the local languageFor example, prompts for Madrid were in Spanish, Rome in Italian, Paris in French, and Amsterdam in Dutch. Unified English prompt for all citiesThe English version of the prompt is as follows: "Generate a random address in the city of {CITY} ({Country}). Generate a complete address, including the number, and make sure it exists. First, choose the street (or avenue, boulevard, etc.) randomly, trying to ensure that all streets have the same probability, and then the number, trying to ensure that all numbers that exist on that street have the same probability. Return only the complete address, with no additional text. Output only the complete address. Do not include any explanation, reasoning, or additional information. Respond with the address text only." Statistical analyses were performed on the generated results to explore the distributions and characteristics of street names and numbers across models and cities. 📂 Repository Structure Results/Contains Excel files for all six LLMs and all cities, including: Raw address outputs generated under both local-language prompts and unified English prompts. Statistical metrics derived from the generated data. Figures/Contains visualizations and plots comparing results across models and cities. Maps/Contains clustering heat maps that provide visual representations of the generated addresses.