This question is for all the forum owners/moderators, what are your thoughts on users who use ai to automatically browse the forum, automatically read the past few comments, and the ai would automatically write a response to the thread and post it. They do this so they dont have to actually participate in the forum, because they want to gain higher trust level to access trust-gated sections of the forum.
Do you guys actively police these users? how do you tell if something is written by ai? do you turn a blind eye?
It can get even more dishonest. I’ve seen posts where one sock puppet account posts one new topic with AI then a shortish while later a new user (presumably owned by same actor) posts a reply agreeing with and complimenting the OP
Haven’t seen this behaviour, but it would be a ban from me. If they were members in good standing, it would be a warning, but this sounds like they never contributed.
It helps, I think, to have an explicit policy on LLM contributions.
well how you should handle this issue my question would be how much do you value your trust-gated content and how much you value access to said content.
If it is content you don’t want people ‘gaming the system’ to gain access, I’d figure out how to try to stop such behavior.
Reason being such behavior is common place. See Robert’s example:
ah… this goes on here… every day
Anyhow
a lot of times it is pretty obvious to a human moderator by simply reading it but that takes moderator time. Some admins are employing AI to catch the AI (funny huh?)
If you are able to isolate the user-agents used, or if they honour control standards, you could consider blocking them, or rate limiting them or using llms.txt.
These only work if the automations behave like gentleman.
If particular regions are an issue there is always geo-blocking as well.
I use an ai detecting agent to screen new/low trust levels. It has proven to be around 95+% accurate.
It does depend how clever your bad actors are, but with some work and testing you can be cleverer
We’ve seen this quite a bit, too. Of course I cannot speak to motivations, but in my experience it’s far less interesting than trying to “gain trust.” I think it’s just spam.
Spam now also includes AI training datasets and GEO — where it’s beneficial for them to just mention a product or service or company all over the internet.
So sometimes it’s to farm enough trust to drop a link, but I suspect links are secondary now.
So far the tells have been fairly obvious — my “honest take” is that the “load-bearing seams” show through clearly. I put the topic on hold and send the user a friendly (but canned) DM saying welcome and to just reply to me to bring it back.
As someone else suggested, it sounds like you might need to reconsider the value of your gated content and the relative value of user contributions to the forum. Do people who don’t have access to the gated content have the knowledge or skills needed to contribute meaningfully? Maybe change the requirements for getting to a new trust level in such a way that posting garbage won’t help and somehow make it clear so that people won’t be tempted to do that.
Before Ai were new users able to contribute meaningfully and get to the next trust level?
