this post was submitted on 25 Aug 2024
6 points (80.0% liked)
Santabot
27 readers
3 users here now
Santabot is an automated moderation tool, designed to reduce moderation load and remove bad actors in a way that is transparent to all and mostly resistant to abuse and evasion.
This is a community devoted to meta discussion for the bot.
founded 4 months ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
You can try. Make a bunch of accounts on one of those instances that doesn't police their signups very well, downvote everything I've ever done with all of those accounts, and see if I get banned. I think it's more difficult to accomplish this than you think.
You posted that everyone in the US should need a nationally verified ID in order to make a social media account, and got dozens of downvotes. You also have some other unpopular opinions. You said Google should be shut down. You're not banned or close to it. Why are you so sure that this tool is going to ban people for expressing unpopular opinions?
I get where you're coming from. It's a valid concern. I think an important part of the answer will be opening up the process, and maybe even taking it out of my hands as the sole proprietor of all the parameters, so it's a community project instead of my project only.
I posted more about this:
https://slrpnk.net/post/13361827
Right.
Sometimes. Not enough to overshadow other positive participation, in most cases. There's a thing that does happen often, where the bulk of what someone says is their unpopular opinion, and they present it with a lot of hostility, so they spend most of their time collecting mostly downvotes. That will get you banned. That, I think, is a feature, not a bug.
It's a big invasion of everyone's privacy for me to lay out all the data in full. Do you want me to break down its judgements about your user, so you can see some details of at least one case? I can do that, either here or over a DM.
I'd like to be able to lay out a more complete picture, too, if you have ideas for how I can break it down without creating drama.
This is completely fair. What I talked about in https://slrpnk.net/post/13361827, spreading the operation of the tool out to the community instead of me operating it only, seems like it could be a good solution.
Like I say, try it. It's not impossible to do, but I would be surprised if anyone could make this work in reality without creating dummy accounts on an industrial scale.
An approach that will work better is to post content that will attract a lot of upvotes, raise your own user's rank, and then downvote everything I've ever done from that single highly-ranked user. You can try that, as an alternative or in conjunction, and see how well it works, if you want to try.
It's definitely not a silver bullet. I have a tendency to look at the whole thing with rose-colored glasses, when it's not perfect. I'm completely open to people poking holes in it or figuring out things that I've missed. All I would ask is that it be based in how it actually behaves, not just a theory about how it's going to do all these terrible things, that's not based on observing how it works in practice.
Then if we look together at what it actually does, and you find a problem, we can agree on it and I can potentially even fix it.
challenge accepted!
I got some questions tho:
My contraversial opinions got me banned by the bot and unbanned recently i assume due to tuning of some parameters of something.
I saw that post i think it will definatly help alleviate some of the concerns. Open sourcing it sounds like a good idea till u realise anyone can then use it to purposely be malicious so thats an even worse idea its a complex issue that thankfully i dont gotta solve.
I do t see how its a privacy violation all the raw votes are public anyways i guess its just ur bots judgements that are private. Ur welcome to post my results publicly since its all open anyways.
I mean u can always anonimise the data use a cheap locally hosted ai to paraphrase/extract the relevent points from the text and hash + salt usernames etc.
I will try it if i can get away with using my own instance then creating accounts on an industrial scale is easily doable. Thx for the info do instances have ranking or can i manufacture heigh ranked accounts on a single instance (even if i cant doing it across many instances is just a matter of purchasing a hacked account list.
Im greatfull for you licence to develop a system designed to try abuse ur bot. As long as i dont get defederated it should be doable without resorting to illegal practice's (buying stolen accounts etc).
I can print about 1k upvotes per 5minutes with 200accounts each account will have about 1/200th that many votes per 5minutes (rank building). Is this pace significant enough to influence ur algorithm?
Currently just proof of concept on my own server (although it is federated). Can easily be defederated to block this but thats only cos i dont wanna be botting someone elses server. Accounts are created with significantly real enough data to feed to an llm/image generator to make them indistinguishable from real account activity to farm votes for rankings or disseminate propaganda or whatever.
All instances are equal.
Not at all. Votes are weighted by the "rank" of the user, as determined by the weight of votes coming in from other users. It's a modified version of PageRank, with the same strengths and a lot of the same vulnerabilities as PageRank.
They're not considered differently in the algorithm. If only upvotes are coming out of any given instance, they're considered to be the same as any other upvote.
What do you mean?
I'm deliberately not providing too much introspection into how things work numerically, because I don't want to provide a roadmap for how to abuse the system. It's impossible to make this type of system bulletproof. All I'm saying is that I think it's a lot more difficult to abuse it than you think it is.
Abusive moderation already exists on Lemmy. Automating it based on the community at large, at this point, sounds to me like more of a blessing than anything else. I don't think anyone who wants to enact abusive moderation would bother with this tool. They would just ban the people they don't like and think no more about it.
I don't mean that it's a privacy violation in the sense of exposing people's votes. What I mean is that I don't want to create a bullying atmosphere, where I'm holding up one person's account and outlining all the reasons they are a bad poster and are banned. I prefer to do that only for people who have signed up to have it done to them.
No problem. I'm interested in researching the question of how well it works, obviously. I may not be very helpful to you, in terms of giving feedback about how it's working, because as I said, I don't want to provide a roadmap. But if you think you are capable to do this, it helps me understand the system and its strengths and weaknesses to have you make the attempt.
I do think that you should worry about being defederated. I don't plan to make any attempt to influence people to defederate you, but I wouldn't be surprised if some administrators take notice of what you're doing and defederate you on their own. Vote-rigging is one of a handful of behaviors which admins feel strongly enough about to defederate for.
I doubt it. You need to outweigh millions of real votes. You can leave your process running and eventually reach that level, but I think that before you start creating an impact on the algorithm, you will be noticed and defederated. I could be wrong, of course, which is why it's a useful experiment.
I think you should also consider the performance implications of your work. If someone starts seeing increased load on their server, and they look into it and discover that you are federating clearly fake votes in the range of thousands per hour and that's why their performance is suffering, they are likely to defederate you and ban your accounts on other instances now and in the future, at the very least. I'm fine with you doing this experiment, but that doesn't extend to server administrators being fine with your impacts on them.
In fact, I recommend that you send a quick note to the server administrator of whichever instance, outlining what you're doing and asking if it is okay from their point of view. I'm not on board for you causing operational problems to anybody.
I spun up my own instance for that very reason as to not be fucking with some poor admin and am only federating with large instances (the ones u post and comment on) to avoid loading any small instances unnessasarilly.
Your karma farm needs to be federated in order to work the way you want it to. It's not enough for your instance to federate with the world, that specific community also needs to.
I think I worked out a way to make it influence the vote bot rankings without polluting a real instance, but you'll need to redo all the votes so they federate out. I think it'll work fine if you just have all the vote bots undo their votes to each other, and then redo them. Give it a try.
Ahh so the karma farm isnt federating? This is my first time running anything federated. If somone from ur instance subscribes to the karma farm community will that then federate it to ur instance and thus allow ur bot to count those votes?
Yes, that's right. Communities are not federated except to instances that are subscribed to them.
Also, it only counts things going forward. When you subscribe to a community, you don't get the votes going back historically. You only get new votes from that point forward.
Yes. I've done a version of that now. If you have the bots in the karma farm make all new posts and mutual votes, then I should be able to pick it up now. The posts you made before won't take effect, because nothing was federating out at the time that you made them.
You are banned again, and have been for a couple of weeks. You accrued a lot of negative rank from the "is (blank) a slur" comments section, but not enough to ban initially. I think generally, you're near the borderline, which is why you drift into getting banned from time to time, as you did when some of your earlier participation aged out of the 30-day window, and it became chiefly that one comments section to go on. I think at the present time, it's mostly that you do not have much participation, outside of that one thread, and so the negative rank from that thread is pushing you over into getting banned.
As with almost everyone who gets banned, it's not your opinions that are the issue. It's your delivery. Here's an example that worked very much against you:
That has nothing to do with the topic at hand. It's just inflammatory for no reason. You didn't answer the question or acknowledge the point that the other person was making. You just tried to continue and inflame the bickering.
You could have posted a calm and reasoned attempt to make the exact same point you were making, acknowledging the question and feeling confident enough in your point of view that you were willing to have it probed and challenged. Instead you opted to ignore the comment, and make a counterpoint while dismissing the question.
Like it says in the FAQ, most of Lemmy is highly tolerant of a wide range of points of view. Like most people, though, they will disapprove across the board of someone who is both unpopular and hostile and argumentative about their unpopular opinion. You can get away with one or the other, but not both.