1

I used GitHubCopilot for code completion for a thesis and I am required to cite all usage of AI. Is there anything that I can cite for using github copilot? Have they released a paper or do I just include a link to their website https://github.com/features/copilot?

3
  • There is no such official publication, white paper etc related to GC. They do have a blogpost about launching of GC, and the documentation. So, in a way you have a freehand to choose any link you want to include in your thesis. Apr 12 at 12:54
  • In this case, I've seen people use a "misc" bibtex entry and give the relevant link
    – Taw
    Apr 12 at 13:12
  • As an example of university guidelines: libguides.mcmaster.ca/cite-gen-ai
    – Ben Bolker
    Apr 14 at 12:32

2 Answers 2

1

First, adhere to any rules from your institution about the thesis. That might mean you have to ask someone who is in charge of administering these rules at your institution, at that no one else on the planet can tell you how to comply with them.

Otherwise, this is not the sort of thing I would typically use an academic citation for where there is an entry in the bibliography. Rather, it's something you refer to in your Methods section. You should not merely state that you used GitHubCopilot but be specific about what you used it for in particular. The principles to follow here are to include enough information that someone redoing your work would be able to trace all your steps. That means mentioning the specific software used, the version of that software, and information necessary for someone to find the software, traditionally the name of the company providing the software and where that company is located (i.e., City, (state/region if relevant), Country). This is the same way that you would typically report any commercially available software or other materials necessary for reproducing your work.

Specialized software created by academics for academics should typically be cited, since citation is the "currency" by which those academics are credited for the work they do, but that's not typically necessary for commercial software authored by a company.

If you must provide it as a citation, refer to the style guide of whatever citation system you are using.

1
  • This seems like an instance of isolated request for rigour. Nobody in computer science says in their publications what IDE they used or what code autcomplete they used, much less the exact versions of these tools. This is because none of this is important to reproduce the results. This is different of course when the tool itself is being studied. For a tool like github copilot, version numbers are especially useless for replication, since future researchers won't have access to past versions of copilot.
    – Polytropos
    Apr 13 at 22:49
1

Like the previous answer has said, follow the guidelines of your institution on how to report copilot use.

As for what to cite, I think a reasonable approach could be to cite the posting where they introduced it, say with which IDE you have used it, and cite in addition the changelog of the applicable plugin, if any. To be sure you are doing it correctly, clear this with your advisor. For copilot specifically, I would also probably cite the Codex paper, as it formally introduced the underlying model family.

In general academic writing, the right place to mention a tool like this would be acknowledgements, I think, where the acknowledgements would go to the developers of the tool, citing a suitable publication, if the tool has been useful (beyond the ordinary background noise of the usefulness of application software in general) in the course of the project while not being needed to replicate the results. Mentioning it in the methods section would seem right to me only if the tool is needed to replicate results (so e.g. a computer algebra tool would be mentioned in the methods section if one needs it to verify a calculation, but might be mentioned only in acknowledgements if it was used but the result can be easily verified by hand).

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .