When the topic of AI submissions flooding open source projects pops up, my immediate reaction is to think "see, this is why you disregard intentions". Because I genuinely believe a lot of the people submitting this slop are trying to help the project, even if in reality they’re harming it, by wasting the maintainers’ time with their crap.They cause harm and deserve to be treated as a source of harm, simple as.
And while most projects could/should use more money, I don’t think that’s the solution; it allows the devs to handle more workload, sure, but the goal should be to reduce it. I think this will be eventually done through pre-sorting contributors: a cathedral for the outsiders, but a bazaar for the insiders.
I honestly do not understand why anyone would ever for any reason submit AI-generated code to any FOSS project.
What’s the motivation behind that kind of behavior? What do people get out of it?
makes a little green box on their github activity chart that they can show to a potential employer
They think the code is good and makes the project better? They want to test/prove their “skills”? Its partnof their day job?
Basically the same reason anyone contributes, just they got sold a bag of rocks and told how amazing a hammet it is.
Seems like a natural next step is some kinda reputation system for a project’s contributors. If you’ve written 50 successfully merged PRs, you’re certainly less likely to make trash in any method. Create a mentorship heirarchy. It sounds very helpful no matter what. Then the people who have merged 0 PRs to the project will likely work harder too, since they know they will be at the most scrutiny
Its trivial for ai to successfully merge 50 prs into the projects of other ai.
It be frustrating for a beginner that every time they try to contribute they are ignored because they have yet to contribute.
Web of trusted peers maybe? Honestly beyond AI not all contributions, no matter how helpful, are equal as a metric. That is to say linus torvolds himself might not be the best guy for contributions for UI changes to Genomic research tool. Clearly VERY reputable on a massive OpenSource project, but not in related work (as far as i know for the stupid example i made up).
This is how they kill open source.
Mitchell Hashimoto is trying to build a reputation system to combat this https://github.com/mitchellh/vouch
