[Request] Openly describe your AI/LLMs usage

[ghgabr]

AI is, sadly, clearly used a lot in the project; can you describe precisely how, either in the README or in a separate .md file?

It would help people decide how much to trust the app.

What I found is about 80 commits co-authored or by Copilot in this repository, 50 in https://github.com/celzero/firestack, 20 in https://github.com/celzero/redir and a handful in the other repositories; but maybe they weren't the only ones where LLMs were heavily used?

I could only find an explicit mention of its usage for the development in #2083 .

[ignoramous]

This app isn't vibe-coded.

The use of AI is mostly that of an automated typewriter, if that makes sense. I am not sure why that needs to go in the README, when "AI" commits are clearly marked with authorship.

[ghgabr]

If it's not vibecoded, describing how AI is used would just prevent people from thinking it is.

And actually, it's not that easy to notice the usage of AI in this project: while the avatar of Claude is well known and easy to spot in GitHub's contributors section, the Copilot one is little known (and with its colors, it doesn't stand out).
I myself only noticed it now, and found out of the copilot's usage with git searches.

There's a wide spectrum of possible ways to use AI, from vibecoding an app to not using it at all; so it's far from pointless to clarify your methodologies, in my opinion.

Your "typewriter" anology doesn't say enough, unfortunately.

Here's a good (if big) checklist of things I'd like to be clarified from the Rethink developers:

• How often, and how much do you use LLMs to help in writing the code?

• Do you always, manually review their output? How carefully?

• Specifically,

  • do you make sure to understand every part of their code, before pushing it to GitHub?

  • Do you check that the AI's code changed everything that needed to be changed?

  • Do you make sure to understand the effects of the changed code to the rest of the software?

  • Do you correct everything that doesn't seem ideal, or sometimes leave some less optimal choices of the LLM? (for example for what concerns the organization of the code, the length of functions, the names of things...)

• If you use them to help write commit messages, do you check those carefully as well?

• Do you do all the above even for contributions from external people, especially when they're openly made with the help of LLMs?

• Only the commits with an AI agent as their author, committer or co-author contain code made with the help of an LLM?

If you do want to answer please strive to be honest with yourselves and the users, very few AI users always review everything carefully.

A lot of questions when I'm not paying you, I know, but this app intercepts all of a device's network traffic, so it's important to understand how it's developed.
And it's probably used by many people, it's even recommended by GrapheneOS.

Finally, some people also have ethical concerns about AI usage; that's another reason to be transparent about it.

[ignoramous]

Thanks. Appreciate your concerns, all of which are valid.

But in the README? That's a bit too much to put in there on just one topic. A blog post / in the docs? Sure. Another .md file? May be.

Any indications of GrapheneOS themselves doing so? Or any other "recommended" project elaborating their "AI use" in README that I can look at to inform what we should perhaps do? Thanks again.

[ghgabr]

I apologize for the very late reply, I've had a few problems.

I'd actually prefer a separate file, on the root of the project; it would still be easy to see, and in the future something like that might be picked up automatically by F-Droid or something else.

I would just briefly link to it in the README.

The only project I can recall to have done anything remotely similar is Aurora Store (https://gitlab.com/AuroraOSS/AuroraStore#ai-usage).

It's not detailed and might not be completely honest, though.

No, for what I can tell GrapheneOS has not done anything like this.

Thank you for considering the proposal.

[luckygitt]

A lot of questions when I'm not paying you

You could always sponsor/ make a donation 😁