Time Nick Message 19:27 sfan5 I feel like the invention of code AIs isn't helping our PR review bottleneck/shortage 19:28 sfan5 there's also a big space of (potential) contributions where "writing the code" is the smallest problem, but API design and fitting it into the roadmap is essential 19:31 Krock I'm honestly not motivated to review AI PRs if the author does not appear to have enough experience to justify the decisions the magic box made for them 19:33 Krock and sure - the roadmap decision is also important such that the API does not blow up in size with functions that aren't generic enough. 19:40 [MatrxMT] Maybe we need a statement? I had another core dev privately complaining about it 19:42 Krock I'd surely welcome a statement like "Prior submitting your PR, you should be able to explain what the changes are supposed to do.". 19:42 [MatrxMT] The fact people can state "The whole PR is AI slop" (#16929) and have their PR reviewed and approved is a bit ridiculous 19:42 ShadowBot https://github.com/luanti-org/luanti/issues/16929 -- Add vector2 lua API for 2D vectors by kromka-chleba 19:43 MTDiscord It is 19:43 rubenwardy See my previous message about having an AI policy 19:43 [MatrxMT] Also "Read unit tests instead of trusting AI" as a to-do bullet point in the same PR 19:43 MTDiscord Honestly, the AI PR makers are vibe coders 19:44 MTDiscord Those, bluntly, aren't skilled programmers who slightly abuse AI agents, but at least know what's going on 19:44 [MatrxMT] Yeah, as long as you use it as a StackOverflow under steroids why not 19:46 MTDiscord Yeah, but if somebody goes and vibe codes the whole thing and avoids taking responsibility for it by saying "it's all AI slop"... 19:48 MTDiscord I'd suggest requiring providing reasons for API design, and at least one reason for each API function. And if somebody does provide them, but they look bad or make little sense or the "author" can't answer questions about it... whole PR should be rejected. 19:49 [MatrxMT] The issue is, I can ask an LLM to write that for me 19:50 MTDiscord Sure, but if reviewer doesn't like the reasons, reject 19:50 MTDiscord Or if reviewer asks a question on those and "author" can't answer it coherently 19:50 MTDiscord If it is coherent and makes sense, it's not an issue 20:00 [MatrxMT] I don't know, I think we're overthinking it, complicating the process for everyone 20:31 MTDiscord I reviewed that PR because I figured after taking a quick look that it was fairly close to what I wanted. There ended up being a bit more discussion (mostly about design) than I anticipated. 20:31 MTDiscord I also read the unit tests and they were not really bad back then, and are fine now. I would hope that the author has also read the unit tests by now..? 20:34 MTDiscord I ultimately approved because the PR is in what I think is a mergeable state. I don't see what I would gain by being picky about how it got there. 20:35 MTDiscord An AI policy sounds like a good idea. Contributors should probably be expected to have fully reviewed (i.e. at least completely read) AI-generated or assisted contributions. 20:37 MTDiscord And well, obviously contributors assume responsibility over the PRs they submit. I don't think kromka-chleba was trying to avoid responsibility here.