10

I keep seeing users like this, the quality of their posts is abysmal:

terrible answer

Can we get a captcha or something? Would that help? What else can be done? One of those things where you have to solve a little puzzle?

Are they using the API? If so, can we lock that down? Or are they simply using something like Selenium? If there is a solution, how can we force SE to implement it?

There is also a trick where you hide form elements and ensure users (even blind users) won't fill them but bots do

I really think its time to brainstorm what can be done to stop the influx of AI generated nonsense. Humans flagging them is not enough.

7
  • 2
    This isn't actually spam, as it does no advertising. AI generated spam is a quite specific thing, which is actual spam that uses AI generated content to help mask that it's spam. While that type of spam is on the rise, this isn't that. If you think something is AI generated content, then you should be raising an "in need of moderator intervention" flag. This might have been intended to get people to look at the user's profile, but there's nothing specific to indicate that and the content which was in the profile is gone now that the profile has been deleted or destroyed.
    – Makyen
    Mar 17 at 23:44
  • 2
    While a captcha might help, it's not going to be all that effective, as most spam appears to be posted by humans. There is a portion of spam that's posted by actual bots. A CAPTCHA might help in those cases, unless the bot just uses a CAPTCHA-solving service. The company's response is a trade-off vs how difficult the company wants to make posting by legitimate users. Given that there are already complaints about too many CAPTCHAs, there's definitely a limit as to how much people will accept prior to just leaving. I know I've recently left other sites due to too much time spent on CAPTCHAs.
    – Makyen
    Mar 17 at 23:51
  • 2
    @Makyen This isn't actually spam False. this isn't that False. most spam appears to be posted by humans False.
    – Gantendo
    Mar 19 at 18:41
  • 1
    @Makyen - The content is in the screenshot absolutely is spam, the content is unacceptably low quality, no doubt attempting to setup a question that can be "answered" with content that will contain an advertisement. It's what we call a spam seed.
    – Ramhound
    Mar 27 at 23:15
  • 1
    @Makyen - Futhermore, that answer even if it's not actually spam, was generated by a LLM. I won't explain HOW I know that, but the level of confidence of that fact, is more than 100%. Since you are a moderator, you know the content was generated by a LLM, I strongly disagree that the content wasn't spam or a spam seed. In the end it does not matter, content that was not helpful, was flagged and properly deleted. The user was going to be destroyed regardless if it was deleted as spam or for submitting low quality ChatGPT junk.
    – Ramhound
    Mar 27 at 23:22
  • Do you think some sort of AI is needed to detect it? /s Mar 28 at 17:50
  • @AndrewMorton Ironically, OpenAI has a pretty cool captcha system with little puzzles you have to solve. If SE implemented something similar but built in-house that might work.
    – Gantendo
    Apr 5 at 5:49

1 Answer 1

9

I'm half certain at least some of these is some poor person in a sweatshop rather than something more sophisticated. It would be nice to have better tools - and we've requested for it

In this case I suspect the goal was to get people to see the profile

Honestly, If I knew what would motivate SE to do something, I'd be heavily leveraging it, but practically we're still 'small' enough that its not a huge deal.

If there's posts, flag them - with a custom flag, and we can deal. We take a dim view of spammers and people posting chatgpt garbage so we're generally happy to review and act appropriately.

If it reaches a silly unsustainable level, it would hopefully be leverage to deal with it, but I somehow doubt the powers that be have that as a priority.

10
  • 1
    In the past, what has worked to get SE to do something the community wants? I am no expert but I read some things about the Monica situation and it sounds like a challenge to get them to implement improvements. We don't want to take hostages or firebomb stuff and that shouldn't be necessary.
    – Gantendo
    Mar 15 at 13:29
  • Probably the only way to know if we are dealing with a sweatshop worker or an unsophisticated bot would be to implement on of those little puzzles, right? Is there another way? Captchas are damn easy to break and it would be trivial for a botwriter to implement a captcha breaking service which would cost them very little.
    – Gantendo
    Mar 15 at 13:31
  • 1
    I wish I knew how to leverage SE, I'd be in a better place to do things :D. We do have captchas BTW, but practically the best we can do is delete them as we catch em. On the bright side, nuking spammers is wildly satisfying, and a good chunk of the network wide moderation teams probably dislike GPT posters and will be quite happy to deal. SU's got a relatively small flag load so custom flags here are probably best
    – Journeyman Geek Mod
    Mar 15 at 13:35
  • In the past, we didn't need to do much much. the management of the time was hands on, it was a smaller network, and the people who ran the network considered the Q&A sites to be of paramount importance.
    – Journeyman Geek Mod
    Mar 15 at 13:38
  • So we all move to codidact where there is a small hands on team, and then after a couple of years codidact has grown so much and has become so bureaucratic and has lost sight of their original mission that we move back to SE. And we keep moving back and forth every few years?
    – Gantendo
    Mar 15 at 13:45
  • Well, or flag and let us deal with it.
    – Journeyman Geek Mod
    Mar 15 at 13:47
  • I could automate the flagging: "if a newly created user creates a profile with a hyperlink and picture and profile text before answering/commenting it is spam". and then you could write a script that says: "if Gantendo flagged it, nuke the user"
    – Gantendo
    Mar 15 at 13:53
  • Which would get tripped by someone creating an account across from another site.
    – Journeyman Geek Mod
    Mar 15 at 13:54
  • True, but that is fixable with a simple check. SE uses something similar to en.wikipedia.org/wiki/Wikipedia:Unified_login so that shouldn't be hard to detect.
    – Gantendo
    Mar 15 at 13:54
  • 1
    Account age is also mentioned on the profile so we can take that into account when flagging suspicious users. Such a behaviour-based detection system is not difficult to implement and a great alternative to a cost-based anti-spam system en.wikipedia.org/wiki/Cost-based_anti-spam_systems
    – Gantendo
    Mar 15 at 14:01

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .