August 3rd, 2021

Today I decided to run through some Project Euler problems just for fun after many years.

As I wrote a function signature for the first problem, to my complete surprise, this happened.

I had forgotten that I had installed Github Copilot a few weeks ago.

I thought it'd be much more interesting to try and get Copilot to answer as many as possible correctly with minimal human intervention.

  1. Copilot is trained on Github, where nearly all Project Euler solutions are hosted.
  2. Human-ML collaboration is the future of programming so it’d be cool to dig into what it looks like.

I wanted to see how far I could get with as little effort as possible and what trouble I would run into.

So I decided to take full assistance of the AI and do the fastest possible method to actually solve the problem, so long as I understood and agreed with the solution.

Here’s what followed.

Problems 1-3: Easy and perfect

The suggestions were readable, iterative and efficient. There may have been more clever ways to do these, but this was the first thing I would've thought of as well.

It even knew what I would run the function on

Powered by Fruition