I have noticed that lemmy so far does not have a lot of fake accounts from bots and AI slop at least from what I can tell. I am wondering how the heck do we keep this community free of that kind of stuff as continuous waves of redditors land here and the platform grows.
EDIT a potential solution:
I have an idea where people can flag a post or a user as a bot and if it’s found out to be a bot the moderators could have some tool where the bot is essentially shadow banned into an inbox that just gets dumped occasionally. I am thinking this because then people creating the bots might not realize their bot has been banned and try and create replacement bots. This could effectively reduce the amount of bots without bot creators realizing it or know if their bots have been blocked or not. The one thing that would also be needed is a way to request being un-bannned if they get hit as a false positive. these would have to be built into lemmy’s moderation tools and I don’t know if any of that exists currently.
My instance has “Rule 3: No AI Slop. This is a platform for humans to interact” and it’s enforced pretty vigorously.
As far as “how”:
Sometimes it’s obvious. In those cases, the posts are removed and the account behind it investigated. If the account has a pattern of it, they get a one way ticket to Ban City
Sometimes they’re not obvious, but the account owner will slip up and admit to it in another post. Found a handful that way, and you guessed it, straight to Ban City.
Sometimes t’s difficult on an individual post level unless there are telltale signs. Typically have to look for patterns in different posts by the same account and account for writing styles. This is more difficult / time consuming, but I’ve caught a few this way (and let some slide that were likely AI generated but not close enough to the threshold to ban).
I hate the consumer AI crap (it has its place, but in every consumer product is not one of them), but sometimes if I’m desperate, I’ll try to get one of them to generate a similar post as one I’m evaluating. If it comes back very close, I’ll assume the post I’m evaluating was AI-generated and remove it while looking at other content by that user, changing their account status to Nina Ban Horn if appropriate.
If an account has a high frequency of posts that seems unorganic, the Eye of Sauron will be upon them.
User reports are extremely helpful as well
I’ve even banned accounts that post legit news articles but use AI to summarize the article in the post body; that violates rule 3 (no AI slop) and Rule 6 (misinformation) since AI has no place near the news.
If you haven’t noticed, this process is quite tedious and absolutely cannot scale under a small team. My suggestion: if something seems AI generated, do the legwork yourself (as described above) and report them; be as descriptive in the report as possible to save the mod/admin quite a bit of work.
That’s interesting I suppose everyone has their own moderation styles. To me I am not 100% opposed to all AI. I define AI slop more like really low effort posts and bulk posts. So a person who is just posting all AI generated content and cross posting to tons of community. Basically AI spam I guess you could say. If someone was to say generate an AI image and make a post talking about the prompt they used and maybe sharing what they like about the image and then commenters make derivatives or share their own results using a similar prompt I could see that sort of post being useful. Maybe there is a balance… but at the same time I can see that some people might prefer an instance that takes more of a hard line stance.
Maybe you could train an AI to help you with the tedious work of banning AI? Lol
Seems to me like we could build a bot for this…🤔
I refuse to reward the “Create problem, sell solution” business model these tech companies are employing lol.
I hear you, but if we write it, it would be more “use their weapons against them”. Bot judo.
All while the environment and electrical grid weeps.
Do you really need that much power for a local model? I haven’t had a team green card in years so idk
Yeah.
The problem is that this is only going to get harder. First of all, AI is going to get better and be able to produce more natural sounding stuff.
But also, people will inevitably get affected by AI as well and people will drift towards sounding more like AI too. So both AI and humans will converge on each other and they’ll likely be impossible to tell apart in general in not too many years.
I’m not sure how we solve this tbh.
The “AI checkers” that schools/unis use has found a strong correlation between neurodiversity and sounding like AI. Basically, AI sounds autistic, so autistic people get flagged as AI.
That instance bans people for nothing, and has some automated ban sync system in place. It’s crazy.
We’re not a general purpose instance, we have a defined mission statement, and the site info clearly states the rules apply to local and federated accounts. 🤷♂️ And the ban syncs are no longer needed as later versions of Lemmy server do the same thing automatically (our automod just implemented something almost identical prior to Lemmy adding that natively).