Generating running routes with GPT-6 Astra and ChatGPT Work
- Source
- simonwillison.net
- Date
Context
Simon Willison reports that OpenAI's ChatGPT Work, running GPT-6 Astra, produced 5K and 10K running loops from a home address using OpenStreetMap data, delivered as an embedded map plus downloadable GPX and GeoJSON route files. The system said it located the address with Nominatim, downloaded nearby roads and trails with Overpass, and calculated the loops locally. That turns a conversational request into an artifact someone can inspect and reuse.
Willison's account also exposes a problem: the map's display code was available, but the Python used to calculate the routes could not be recovered afterward. Willison suspected the conversation had been compacted, which would explain the missing code, though the cause was not established.
- determinism — Whether the same input reliably produces the same output — something LLM systems mostly lack, which changes how you test and debug them.
- context compaction — Summarizing an agent's earlier conversation to free room in the context window so a long session can keep going.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…