Generating running routes with GPT-6 Astra and ChatGPT Work (simonwillison.net)

In a recent demonstration, a user leveraged ChatGPT Work alongside GPT-6 Astra to generate 5K and 10K running routes from their home using OpenStreetMap (OSM) data. The AI processed the request in about 27 minutes, yielding both an embedded visualization and downloadable files in GPX and GeoJSON formats. The generation involved utilizing Nominatim for address location and Overpass to fetch local OSM roads and trails, showcasing the capability of AI in handling real-world applications like fitness routing. Significantly, this showcases the potential of advanced LLMs in combining geographical data with user customization, paving the way for novel applications in fitness, navigation, and outdoor activities. However, the user expressed concerns regarding the lack of transparency in the code execution within the ChatGPT interface, highlighting an area for improvement in AI systems—namely ensuring accessibility to the underlying code and processes used. This raises questions about the ethical implications and reliability of AI-generated outputs in practical scenarios, emphasizing the need for enhanced transparency and user agency in AI interactions.