07-19, 14:30–14:50 (Poland), Plenary Auditorium
An overview of attempts to use large language models (LLMs) within the context of OpenStreetMap (OSM), followed by an exploration of results from experimentation in using OpenAI’s ChatGPT API as an assistant for repairing invalid tags, using the “opening_hours” tag as a practical example.
It is no secret that there has been a gold rush around large language models (LLMs) since the release of ChatGPT in 2022. We are currently in a “hype wave”, as individuals and companies scramble to find ways to get the best use of it. OpenStreetMap (OSM) is no exception, as there have been attempts to find ways to work LLMs into the ecosystem. In this talk I will give an overview of existing attempts of using LLMs within the context of OSM, as well as share observations, both positive and negative, in making use of OpenAI’s ChatGPT API as an assistant for repairing invalid tag values, using the “opening_hours” tag as a practical example.
I was introduced to OpenStreetMap in the early 2010s as a Bachelor's student in Geography. Since then I have completed a Master's degree in Geomatics and work as a research fellow and GIS developer at the European Institute for Energy Research in Karlsruhe, Germany. I work with OpenStreetMap data frequently and have a great interest in analyzing it as a bountiful data source.