Generating running routes with GPT-6 Astra and ChatGPT Work
· Source: Simon Willison
Simon Willison explained how he used ChatGPT Work with the GPT‑6 Astra model (also known as Max) to request 5‑km and 10‑km running routes that start and finish at his home, using OpenStreetMap data. Within half an hour the agent produced an embedded visualization as well as downloadable GPX and GeoJSON files. The 5‑km route shown passes through the port of Granada and several local boulevards, with map information supplied by OSM contributors.
When asked about the process, the model described using Nominatim to locate the address and Overpass to pull nearby roads and trails, calculating loops locally. Willison noted that the Python code executed was not accessible in the ChatGPT interface, which he sees as a lack of transparency. Additionally, when he tried to retrieve the script, the conversation had already been compressed and the model could not recover it, suggesting that systems employing compression should preserve the original text and allow it to be queried via agent tools.
The visualization was created with the “visualize” capability, which generated an HTML file hosted in the workspace and embedded it directly in the chat window. Willison shared the HTML code via a gist so others can review it.
This demonstration highlights the potential of LLMs to generate customized routes from geospatial data, while also underscoring the need for greater openness in code execution—an essential factor for users who must reproduce or audit results.
Read the original article on Simon Willison
This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.