Encourage Using GitHub Copilot to Increase Julia Usage

I think the Julia community should actively encourage the use of GitHub Copilot to strengthen Julia, especially programmers from other languages. Copilot can shorten the learning curve of the language and speed up development in Julia. Currently, Copilot is free for users with an .edu email as well as maintainers of open-source projects. Those two demographics have the strongest influence on Julia’s ongoing success.

If you’re not familiar with the power of Copilot already, it’s amazing because it can:

  1. Interpret other people’s Julia code into plain English comments
  2. Generate multiple Julia code suggestions based on comments in plain English
  3. Generate multiple Julia code completion suggestions based on context clues from the rest of the code in your project as well as your past coding in Julia.

The more Julia users train Copilot’s ai model, the better Copilot will get at at training Julia users. The compounding nature of this will benefit the Julia language and it’s community exponentially.

The next step is would be for the Julia community to create its own free large language model based plugin/extension for code generation/interpretation/completion that specializes in Julia and it’s integrations. It would be good to find a way to integrate this where Julia is most used already, like Juno and Jupyter, but also where it’s not, such as VS Code and Pycharm.

6 Likes

Well, this is not exactly true. It is free only for maintainers of POPULAR open source projects, and it is not so clear to me how this is defined…

1 Like

.edu is not a thing in many other countries than US as far as I’m aware.

3 Likes

Good point.

My last point about the community creating its own open source Julia focused co-pilot would be best but that could take a year. Utilizing co-pilot to help keep Julia’s growth rate up is a temporary solution.

Turns out a .edu email isn’t what they use for verification.

1 Like

The links that you posted are dead. :frowning:

1 Like

Fixed

1 Like

I would strongly advise against using Copilot.
The reason has been stated very cogently here:

13 Likes

IMO “resistance is futile.”

The new Bloom open source ai model could be good for prototyping a free solution.

Many people who resisted evil were told the same.

7 Likes

Yeah, you’re right. We should probably focus on solutions similar to the Polycoder ai model

Today I played with GitHub Copilot more for scripting in Julia. I’m really impressed by how good it’s suggestions are. It’s a little bit better at Python l, but the difference is nowhere near what you’d expect given the difference in the amount of training data.

I have two guesses at why it’s so surprising good at translating English pseudo code to Julia and visa versa:

  1. Julia is a little more similar to English pseudo code than Python

  2. Julia is so similar to Python

I am new to Julia and Copilot sometimes knows what I should be doing when I do not.

2 Likes

Coming into this conversation late (been away from programming in general for a while), but I saw @PetrKryslUCSD 's post about Copilot and I think that’s something to be taken seriously. Drew DeVault offers (I think) reasonable suggestions for how to resolve the concerns. Unless the new version of Copilot resolves these, it seems that a true open source “copilot” is a better recourse. Does anyone know if such a project is in the works, Julia-based or otherwise?

1 Like