Wikipedia editors adopt a policy giving admins the authority to quickly delete AI-generated articles that meet certain criteria, like incorrect citations
Technology
111
Beiträge
46
Kommentatoren
2
Aufrufe
-
Do you think these people surreptitiously submitting articles written by AI are gonna be capable of validating what they're submitting is even true? Particularly if the (presumably effective) Wikipedia defense for this is detecting made up citations?
This kind of thing makes something valuable to everyone, like Wikipedia, ultimately a less valuable resource, and should be resisted and rejected by anyone with their head screwed on
Oh, I think this is a good move by Wikipedia. I just hate to imagine the disaster that ouroboros of AI citing AI generated Wikipedia articles would come up with.
-
Did you generate this comment with a LLM for irony?
It always feels weird when people write an essay as if this is their final quarter project for high school. Too neat, thoughts too organized, much flowery proses.
-
If anyone has specific questions about this, let me know, and I can probably answer them. Hopefully I can be to Lemmy and Wikimedia what Unidan was to Reddit and ecology before he crashed out over jackdaws and got exposed for vote fraud.
Well now I want to know about jackdaws and voter fraud
-
I've started to drop using emdashes because AI ruined them--bastards.
-
This post did not contain any content.
It's a step. Why wouldn't they default to not accepting any AI generated content, and maybe have a manual approval process? It would both protect the content and discourage LLM uses where llms suck.
-
Well now I want to know about jackdaws and voter fraud
unzips
-
If anyone has specific questions about this, let me know, and I can probably answer them. Hopefully I can be to Lemmy and Wikimedia what Unidan was to Reddit and ecology before he crashed out over jackdaws and got exposed for vote fraud.
How frequently are images generated/modified by diffusion models uploaded to Wikimedia Commons? I can wrap my head around evaluating cited sources for notability, but I don't know where to start determining the repute of photographs. So many images Wikipedia articles use are taken by seemingly random people not associated with any organization.
-
If anyone has specific questions about this, let me know, and I can probably answer them. Hopefully I can be to Lemmy and Wikimedia what Unidan was to Reddit and ecology before he crashed out over jackdaws and got exposed for vote fraud.
Is there a danger that unscrupulous actors will try and build out a Wikipedia edit history with this and try to mass skew articles with propaganda using their "trusted" accounts?
Or what might be the goal here? Is it just stupid and bored people?
-
How frequently are images generated/modified by diffusion models uploaded to Wikimedia Commons? I can wrap my head around evaluating cited sources for notability, but I don't know where to start determining the repute of photographs. So many images Wikipedia articles use are taken by seemingly random people not associated with any organization.
So far, I haven't seen all that many, and the ones that are are very obvious like a very glossy crab at the beach wearing a Santa Claus hat. I definitely have yet to see one that's undisclosed, let alone actively disguising itself. I also have yet to see someone try using an AI-generated image on Wikipedia. The process of disclaiming generative AI usage is trivialized in the upload process with an obvious checkbox, so the only incentive not to is straight-up lying.
I can't say how much this will be an issue in the future or what good steps are to finding and eliminating it should it become one.
-
Is there a danger that unscrupulous actors will try and build out a Wikipedia edit history with this and try to mass skew articles with propaganda using their "trusted" accounts?
Or what might be the goal here? Is it just stupid and bored people?
So Wikipedia has three methods for deleting an article:
- Proposed deletion (PROD
An editor tags an article explaining why they think it should be uncontroversially deleted. After seven days, an administrator will take a look and decide if they agree. Proposed deletion of an article can only be done once, even this can be removed by anyone passing by who disagrees with it, and an article deleted via PROD can be recreated at any time. - Articles for deletion (AfD A discussion is held to delete an article. Pretty much always, this is about the subject's notability. After the discussion (a week by default), a closer (almost always an administrator, especially for contentious discussions) will evaluate the merits of the arguments made and see if a consensus has been reached to e.g. delete, keep, redirect, or merge. Articles deleted via discussion cannot be recreated until they've satisfied the concerns of said discussion, else they can be summarily re-deleted.
- Speedy deletion: An article is so fundamentally flawed that it should be summarily deleted at best or needs to be deleted as soon as possible at worst. The nominating editor will choose one or more of the criteria for speedy deletion (CSD), and an administrator will delete the article if they agree. Like a PROD, articles deleted this way can be recreated at any time.
This new criterion has nothing to do with preempting the kind of trust building you described. The editor who made it will not be treated any differently than without this criterion. It's there so editors don't have to deal with the bullshit asymmetry principle and comb through everything to make sure it's verifiable. Sometimes editors will make these LLM-generated articles because they think they're helping but don't know how to do it themselves, sometimes it's for some bizarre agenda (e.g. there's a sockpuppet editor who's been occasionally popping up trying to push articles generated by an LLM about the Afghan–Mughal Wars), but whatever the reason, it just does nothing but waste other editors' time and can be effectively considered unverified. All this criterion does is expedite the process of purging their bullshit.
I'd argue meticulously building trust to push an agenda isn't a prevalent problem on Wikipedia, but that's a very different discussion.
- Proposed deletion (PROD
-
Well now I want to know about jackdaws and voter fraud
-
Even saying fair question set off alarms. At this point saying anything good about a response at the start is immediate red flag.
These lists of red flags make me feel like I must be a replicant. I wrote a comment just like that one, em dash and all, on a different site just the other day, with my own organic brain!
My first instinct was to use an em dash instead of that last comma, but it seemed too on the nose.
-
This post did not contain any content.
I downloaded the entirety of wikipedia as of 2024 to use as a reference for "truth" in the post-slop world. Maybe I should grab the 2022 version as well just in case...
-
reddit allows GOOGLE to scrape it for its AI, because google allows them to use thier v3captcha for thier moderation and banning purposes.
-
If anyone has specific questions about this, let me know, and I can probably answer them. Hopefully I can be to Lemmy and Wikimedia what Unidan was to Reddit and ecology before he crashed out over jackdaws and got exposed for vote fraud.
Unidan was a legend, he will be missed.
-
So Wikipedia has three methods for deleting an article:
- Proposed deletion (PROD An editor tags an article explaining why they think it should be uncontroversially deleted. After seven days, an administrator will take a look and decide if they agree. Proposed deletion of an article can only be done once, even this can be removed by anyone passing by who disagrees with it, and an article deleted via PROD can be recreated at any time.
- Articles for deletion (AfD A discussion is held to delete an article. Pretty much always, this is about the subject's notability. After the discussion (a week by default), a closer (almost always an administrator, especially for contentious discussions) will evaluate the merits of the arguments made and see if a consensus has been reached to e.g. delete, keep, redirect, or merge. Articles deleted via discussion cannot be recreated until they've satisfied the concerns of said discussion, else they can be summarily re-deleted.
- Speedy deletion: An article is so fundamentally flawed that it should be summarily deleted at best or needs to be deleted as soon as possible at worst. The nominating editor will choose one or more of the criteria for speedy deletion (CSD), and an administrator will delete the article if they agree. Like a PROD, articles deleted this way can be recreated at any time.
This new criterion has nothing to do with preempting the kind of trust building you described. The editor who made it will not be treated any differently than without this criterion. It's there so editors don't have to deal with the bullshit asymmetry principle and comb through everything to make sure it's verifiable. Sometimes editors will make these LLM-generated articles because they think they're helping but don't know how to do it themselves, sometimes it's for some bizarre agenda (e.g. there's a sockpuppet editor who's been occasionally popping up trying to push articles generated by an LLM about the Afghan–Mughal Wars), but whatever the reason, it just does nothing but waste other editors' time and can be effectively considered unverified. All this criterion does is expedite the process of purging their bullshit.
I'd argue meticulously building trust to push an agenda isn't a prevalent problem on Wikipedia, but that's a very different discussion.
Thank you for your answer, I really feel happy that Wikipedia is safe then. Stuff happening nowadays makes me always think of the worst.
Do you think your problem is similar to open-source developers fighting AI pull requests? There it was theorised that some people try to train their models by making them submit code changes and abuse the maintainers' time and effort to get training data.
Is it possible that this is an effort to steal work from Wikipedia editors to get you to train their AI models?
- Proposed deletion (PROD
-
This post did not contain any content.
Wikipedia certainly doesn't need AI to fuck up their articles.
Plenty of biased, incorrect stuff done by themselves. -
Thank you for your answer, I really feel happy that Wikipedia is safe then. Stuff happening nowadays makes me always think of the worst.
Do you think your problem is similar to open-source developers fighting AI pull requests? There it was theorised that some people try to train their models by making them submit code changes and abuse the maintainers' time and effort to get training data.
Is it possible that this is an effort to steal work from Wikipedia editors to get you to train their AI models?
Is it possible that this is an effort to steal work from Wikipedia editors to get you to train their AI models?
I can't definitively say "no", but I've seen no evidence of this at all.
-
So far, I haven't seen all that many, and the ones that are are very obvious like a very glossy crab at the beach wearing a Santa Claus hat. I definitely have yet to see one that's undisclosed, let alone actively disguising itself. I also have yet to see someone try using an AI-generated image on Wikipedia. The process of disclaiming generative AI usage is trivialized in the upload process with an obvious checkbox, so the only incentive not to is straight-up lying.
I can't say how much this will be an issue in the future or what good steps are to finding and eliminating it should it become one.
-
Wikipedia certainly doesn't need AI to fuck up their articles.
Plenty of biased, incorrect stuff done by themselves.Link the incorrect stuff
