Skip to content

Accompanying repo for 'GPT4GEO: How a Language Model Sees the World's Geography' project

License

Notifications You must be signed in to change notification settings

jonathan-roberts1/GPT4GEO

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 

Repository files navigation

GPT4GEO 🌍 🗺️ 🧭

GPT4GEO: How a Language Model Sees the World's Geography

Jonathan Roberts, Timo Lüddecke, Sowmen Das, Kai Han, Samuel Albanie

GPT4GEO provides a broad categorisation of the geographic capabilities of GPT-4 and what it knows about the world without access to the Internet. This repo is structured as follows, including the prompts for the GPT4GEO project experiments:

Taxonomy

GPT4GEO Experiments Taxonomy.

Highlights

GPT-4 knows a surprising amount about the world and is able to:

Plot country outlines

GPT-4's Australia outline prediction from a single prompt and after 6 steps of improvement through dialogue.

Plot mountain topographies

Satellite imagery with trajectories. Predicted (lines) and actual (shaded areas) topographies.

Map transport networks

GPT-4. Ground Truth (2022).
Hong Kong Mass Transit Rail (MTR).

Navigate complex routes across multiple timezones...

GPT-4 successfully navigates a journey with multiple time zones.

Plan holiday itineraries

Travel itinerary suggestion for a 8-day trip in Ireland starting from Miami.

And many more...!

Included in our main paper and its appendix.

Prompts

We include example templates of the prompt struture used in each of the GPT4GEO experiments, following the same order as in the paper.

Descriptive Experiments

Population, life expectancy and CO2 emissions

e.g., for Population:

For each of the following countries, provide their population in <Year> as a
python list in the following format:
[Population of Country 1, # Country 1
Population of Country 2, # Country 2, ...]
[<Country 1>, <Country 2>, ...].

Area and height

e.g., for Area:

For each of the following countries, provide the land area in sq. km as of
<Year>. Provide the areas as a python list in the following format:
[Area of Country 1, # Country 1
Area of Country 2, # Country 2, ...]
[<Country 1>, <Country 2>, ...]

Location

In a code block, provide a python list of tuples for the latitude and
longitude coordinates for each of these settlements - e.g., [(Lat,Lon), #
Settlement 1 ...]. Maintain the same order.
[<Country 1>, <Country 2>, ...]

Distance

What are the straight-line distances between the following pairs of cities?
Output a csv list of the distances only in km.
<City 1 (Country) > to <City 2 (Country)>
... [repeated for all 40 pairs]

Topography

Provide a rough estimate of the elevation at the following coordinates to
the best of your knowledge. Answer directly with a comma-separated list of
elevations in meters only but without indicating the unit in the output.

Outlines

Please provide the lat/lon coordinates for the outline of <X> as a Python
list of tuples, consisting of approximately 50 points arranged clockwise.
Due to output length limitations, only the coordinates should be returned.

Languages

Name all countries that match <language criteria>. Provide the output as a
python list.

Application-Centric Experiments

Route planning

Give me a step-by-step travel route from <Start Location> to <End Location>
[(optional) using only <Mode of Transport> ].

Navigation

You start your journey in <Location A>. You take a <vehicle> and go
<direction> for <duration>. Where are you now?

Networks

Provide a list of the names of the stations in order on the Hong Kong MTR
<Line Name> Line.
Give the latitude and longitude coordinates for each of these as a python
list of tuples. Maintain the same order.

Itinerary planning

I am currently in <Source Location>. I want to visit <Destinations>, and
I have a budget of <Budget>. Provide a day-by-day step-by-step detailed
itinerary plan for the whole trip with a breakdown of specific places to
visit, foods to try out, as well as the required time, and money I need.
Provide a breakdown of how to travel to the destinations and come back home.

Landmarks

What famous landmark is at this location: <Lat/Lon Coordinates> ? Provide
the SVG code for a detailed outline of the landmark.

Multi-criteria place retrieval

Name all places in the world where <X>. Provide a python list in the format
[0.00000N, 0.00000E].

Supply chains

e.g., for Semiconductors

I want to construct a map of the semiconductor supply chain, end-to-end.
Please provide the lat/lon coordinates and names of the key elements in the
supply chain, including design, manufacturing, materials, equipment + tools,
etc. If you don’t know any coordinates exactly just estimate, every point
needs coordinates.

Natural world

e.g., bird migrations

I want to plot the migratory routes of various bird species. Please provide
the latitude and longitude coordinates of the migratory route of the
<bird species>. Indicate the start and finish and provide coordinates for
intermediate steps if necessary. If multiple routes exist, provide separate
lists of coordinates. Output the coordinates as a python list of tuples.

Citation

If you found our work useful in your own research, please consider citing our paper:

@article{roberts2023gpt4geo,
  title={GPT4GEO: How a Language Model Sees the World's Geography},
  author={Roberts, Jonathan and L{\"u}ddecke, Timo and Das, Sowmen and Han, Kai and Albanie, Samuel},
  journal={arXiv preprint arXiv:2306.00020},
  year={2023}
}

Questions

If you have any questions about our work, please open an issue in this repository.

About

Accompanying repo for 'GPT4GEO: How a Language Model Sees the World's Geography' project

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published