ChatGeoPT: Exploring the future of talking to our maps
summary
A toy geospatial assistant called ChatGeoPT translates natural language into valid Overpass API queries, sends them, then feeds the XML results back into GPT-3 to summarize and remember conversation history; the prototype is about 150 lines of code. It successfully refines queries (example: narrowing restaurants in Hereford to Indian cuisine), fails when Overpass returns huge result sets, and surfaces both practical promise and injection/scale risks while pointing toward earth embeddings and an experimental Earth Index.
AI-written summary · about
chapters (13)
- 0:00Introduction to talk
- 0:22AI and maps vision
- 1:04Why we love AI and Maps
- 3:14Bridging AI with geospatial tools
- 5:47Building ChatGeoPT assistant
- 8:18Demo of ChatGeoPT in action
- 10:26Key findings from ChatGeoPT
- 10:46Future of AI in geospatial tech
- 12:52AI powering OpenStreetMap stack
- 13:56How OpenStreetMap can shape AI
- 14:41Earth embeddings concept
- 15:02Demo of Earth Index
- 15:22Closing and resources
related talks
01h 16mChatGPT - and How we Can Benefit from it
44mExploring ChatGPT for Improved Observability
13mGetting Started with Cloud Optimized & What Happens When a Mapion Operator Meets ChatGPT
25mChatGPT Unleashed: From Code Wizards To Silly Shenanigans!
01minteresting experience with ChatGPT
51mA ChatGPT on your own data: how does it work?
46mConversations with ChatGPT: illusion or reality?
01h 01mHow ChatGPT Works