https://github.com/sanand0/antipodes

Show what's on the other side of the Earth. Pure LLM code by O3
https://github.com/sanand0/antipodes

data-visualization llm

Last synced: 11 months ago
JSON representation

Show what's on the other side of the Earth. Pure LLM code by O3

• Host: GitHub
• URL: https://github.com/sanand0/antipodes
• Owner: sanand0
• License: mit
• Created: 2025-04-19T07:49:24.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2025-04-19T08:15:43.000Z (about 1 year ago)
• Last Synced: 2025-05-08T23:57:44.517Z (about 1 year ago)
• Topics: data-visualization, llm
• Language: Python
• Homepage: https://geojson.io/#data=data:text/x-url,https%3A%2F%2Fraw.githubusercontent.com%2Fsanand0%2Fantipodes%2Frefs%2Fheads%2Fmain%2Fantipodal_ocean.geojson
• Size: 307 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Antipodal countries

David McCandless shared a [post](https://www.linkedin.com/posts/david-mccandless-4641b54_which-country-is-across-the-ocean-activity-7317621545374265344-V303): "Which country is across the ocean?"

![Visualization of which country is across the ocean - Eastern Hemisphere](across-east.webp)

This was cool! It shows which country you would reach if you traveled east or west across the ocean in a straight line.

I was curious which country you would reach if you tunneled below in a straight line. This is called the [antipode](https://en.wikipedia.org/wiki/Antipodes) and there are several visualizations of it.

![Antipodes visualization](https://upload.wikimedia.org/wikipedia/commons/6/64/Antipodes_LAEA.png)

I wanted an interactive one. Since LLMs are _so_ good, and O3 was just released, [I asked O3](https://chatgpt.com/share/6803528e-3a04-800c-a8df-8693aa1714d5):

> Write a program that shows each country projected to the opposite side of the globe, but only the parts that overlap the ocean.

>

> Figure out where to get the geospatial data for countries online and get it yourself

>

> I don't need you to visualize it. I need the output in an open format that can be visualized in popular open-source or online geo-tools where I can rotate the globe, click on the countries' opposites and see which country is where.

>

> At the end allow me to download the file and tell me where I can view it.

The attempt was not bad at all, but there was a problem:

![Countries straddling the prime meridian had errors.](error.webp)

So, I started again, using ChatGPT's own prompt suggestion, and adding my description of the error. [Here's the chat](https://chatgpt.com/share/68034776-8cec-800c-a85b-7d6bc94411c0):

> Build a _single_ GeoJSON (EPSG:4326) that shows, for each country, only the parts of its antipode that lie over ocean.

> Carefully handle countries that straddle the prime meridian - UK, France, Algeria, etc.

The result is this [`antipodes.py`](antipodes.py) and [`antipodal_ocean.geojson`](antipodal_ocean.geojson) that it generates.

Click the link below to [view the interactive visualization on geojson.io](https://geojson.io/#data=data:text/x-url,https%3A%2F%2Fraw.githubusercontent.com%2Fsanand0%2Fantipodes%2Frefs%2Fheads%2Fmain%2Fantipodal_ocean.geojson):

[![View the file on a map](result.webp)](https://geojson.io/#data=data:text/x-url,https%3A%2F%2Fraw.githubusercontent.com%2Fsanand0%2Fantipodes%2Frefs%2Fheads%2Fmain%2Fantipodal_ocean.geojson)

I didn't change O3's script except to add the `uv` dependencies and switch from the 1:10m to 1:110m maps for speed.

To run the script:

```bash

wget https://naciscdn.org/naturalearth/110m/cultural/ne_110m_admin_0_countries.zip

unzip ne_110m_admin_0_countries.zip -d ne_110m_admin_0_countries

uv run antipodes.py

```

A few things I learnt from this:

**Ask for the output, not the code**. Models like [O3](https://platform.openai.com/docs/models/o3) and [O4 Mini](https://platform.openai.com/docs/models/o4-mini) can run code, search, reason, read images, etc. while thinking. The models _are_ the environment, now! Let's stop asking for code and ask for the result instead.

**Edge cases are everywhere**. If all goes well, you get AI-speed results. But when something goes wrong (because UK, France, Algeria, etc. straddle the prime meridian, for example), fixing it takes an expert and human-speed results. In fact, the current output is far from perfect (e.g. Indonesia appears in Brazil which is clearly not an ocean.) Programmers under-estimate edge cases, so compensate for this.

**Geospatial analysis is mature**. I was surprised to see `.difference()` in shapes, which made the core part of the code (`build_antipodal_ocean()`) very concise. Libraries like [`fiona`](https://pypi.org/project/fiona) and [`shapely`](https://pypi.org/project/shapely/) have made a big impact.