We're not handling the LLM generative bullshit bots now, anywhere. There's a thing called the dead Internet theory. Essentially most of the traffic on the Internet now is bots.
this post was submitted on 16 Oct 2024
64 points (97.1% liked)
Fediverse
28732 readers
202 users here now
A community to talk about the Fediverse and all it's related services using ActivityPub (Mastodon, Lemmy, KBin, etc).
If you wanted to get help with moderating your own community then head over to !moderators@lemmy.world!
Rules
- Posts must be on topic.
- Be respectful of others.
- Cite the sources used for graphs and other statistics.
- Follow the general Lemmy.world rules.
Learn more at these websites: Join The Fediverse Wiki, Fediverse.info, Wikipedia Page, The Federation Info (Stats), FediDB (Stats), Sub Rehab (Reddit Migration), Search Lemmy
founded 2 years ago
MODERATORS
It’s not just the internet. For example, students are handing in essays straight from ChatGPT. Uni scanners flag it and the students may fail. But there is no good evidence either side, the uni side detection is unreliable (and unlikely to improve on false positives, or negatives for that matter) and it’s hard for the student to prove they did not use an LLM. Job seekers send in LLM generated letters. Consultants probably give LLM based reports to clients. We’re doomed.
Hardly. Just do away with coursework and stick to in-person exams and orals.
Spoken by someone who has never felt with a learning dissability
You can still have extra allotted time, or be provided a wiped computer or tablet. Colleges dealt with these disabilities before llms
I don’t disagree, but it’s probably not that easy. Universities in my country don’t have the resources anymore to do many orals, and depending on the subject exams don’t test the same skills as coursework.
Not even the biggest tech companies have an answer sadly... There are bots everywhere and social media is failing to stop them. The only reason there aren't more bots in the Fediverse is because we're not a big enough target for them to care (though we do have occasional bot spam).
I guess the plan is to wait until there's an actual way to detect bots and deal with them.
Not even the biggest tech companies have an answer sadly…
They do have an answer: add friction. Add paywalls, require proof of identity, start using client-signed certificates which needs to be validated by a trusted party, etc.
Their problem is that these answers affect their bottom line.
I think (hope?) we actually get to the point where bots become so ubiquitous that the whole internet will become some type of Dark Forest and people will be forced to learn how to deal with technology properly.
Their problem is that these answers affect their bottom line.
It's more complicated than that. Adding friction and paywalls will quickly kill their userbase, requiring a proof of identity or tracking users is a privacy disaster and I'm sure many people (especially here) would outright refuse to give IDs to companies.
They're more like a compromise than a real solution. Even then, they're probably not foolproof and bots will still manage.
requiring a proof of identity or tracking users is a privacy disaster and I'm sure many people (especially here) would outright refuse to give IDs to companies.
The Blockchain/web3/Cypherpunk crowd already developed solutions for that. ZK-proofs allow you to confirm one's identity without having to reveal it to public and make it impossible to correlate with other proofs.
Add other things like reputation-based systems based on Web-Of-Trust, and we can go a long way to get rid of bots, or at least make them as harmless as email spam is nowadays.
It's unfortunate that there's such a powerful knee-jerk prejudice against blockchain technology these days that perfectly good solutions are sitting right there in front of us but can't be used because they have an association with the dreaded scarlet letters "NFT."
I don't like or trust NFT's and honestly, I don't think anybody else should for the most part. I feel the same about a lot of new crypto. But I don't necessarily distrust blockchain because of that. I think it has its own set of problems, in that where the record is kept is important and therefore a target. We already have problems with leaks of PII. Any blockchain database that stores the data to ID people will be a target too.
ZK-proofs
This is a solution in the same way that PGP-keys are a solution. There's a big gulf between the theory and implementation.
load more comments
(1 replies)
I saw a comment the other day saying that "the line between the most advanced bot and the least talkative human is getting more and more thinner"
Which made me think: what if bots are setup to pretend to be actual users? With a fake life that they could talk about, fake anecdotes, fake hobbies, fake jokes but everything would seem legit and consistent. That would be pretty weird, but probably impossible to detect.
And then when that roleplaying bot once in a while recommends a product, you would probably trust them, after all they gave you advice for your cat last week.
Not sure what to do in that scenario, really
I've just accepted that if a bot interaction has the same impact on me as someone who is making up a fictional backstory, I'm not really worried wheter it is a bot or not. A bot shilling for Musk or a person shilling for Musk because they bought the hype are basically the same thing.
In my opinion the main problem with bots is not individual acccounts pretending to be people, but the damage they can do en masse through a firehose of spam posts, comments, and manipulating engagement mechanics like up/down votes. At that point there is no need for an individual account to be convincing because it is lost in the sea of trash.
Even more problematic are entire communities made out of astroturfing bots. This kind of stuff is increasingly easy and cheap to set up and will fool most people looking for advise online.
Maybe we should look for ways of tracking coordinated behaviour. Like a definition I've heard for social media propaganda is "coordinated inauthentic behaviour" and while I don't think it's possible to determine if a user is being authentic or not, it should be possible to see if there is consistent behaviour between different kind of users and what they are coordinating on.
Edit: Because all bots do have purpose eventually and that should be visible.
Edit2: Eww realized the term came from Meta. If someone has a better term I will use that instead.
You're missing the big impact here which is that bots can shift public opinion in mass which affects you directly.
Gone are the days where individuals have their own opinions instead today opinions are just osmosised through social media.
And if social media is essentially just a message bought by whoever can pay for the biggest bot farm, then anyone who thinks for themselves and wants to push back immediately becomes the enemy of everyone else.
This is not a future that you want.
I didn't miss it, since my entire post is about manipulation and the second paragraph is about scale.
A bot shilling for Musk or a person shilling for Musk because they bought the hype are basically the same thing.
It's the scale that changes. One bot can be replicated much easier than a human shill.
So my second paragraph...
Lemmy has no capability to handle non-advanced bots from yesteryear.
It's most definitely not capable of handing bots today and is absolutely unprepared for handling bots tomorrow.
The fediverse is honestly just pending the necessary popularity in order to be turned into bot slop with no controls.
Does lemmy and other fediverse stuff currently have such a huge bot problem?
Hard to say. That’s the problem.
A detectable bot problem is a solvable bot problem.
Yes. But at least with the admin group I'm part of, it's dealt with fairly quickly, because we employ automated tools to help fight the spam.
You can't really tell.
It's also a very very VERY small platform compared to other social media platforms like Reddit. (I had another comment where I calculated this but it's ridiculously small)
It is unlikely that it would see anywhere near the same level of dedicated bot activity due to the low return on invested effort.
This is a problem that will become greater once the value of astroturfing and shifting opinion on Lemmy is high enough.
The Fediverse has the advantage of being able to control its size. If 10 million people join lemmy tomorrow and most of them go to lemmy.World and then lemmy.World users start causing trouble then that instance gets defederated.
Other than that we only have human moderation which can be overwhelmed.
We also have auto moderators. The recent spam wave didn't occur on my instance at all. But my Matrix notification channel sure did explode with messages of bots being banned.
I think smarter people than me will have to figure it out and even then it's going to be a war of escalation. Ban the bots, build better bots, back and forth back and forth.
Some news sites had an interesting take on comments sections. Before you could comment on an article, you had to correctly answer a 5 question quiz proving you actually read it.
But AI can do that now too.
Some news sites had an interesting take on comments sections. Before you could comment on an article, you had to correctly answer a 5 question quiz proving you actually read it.
It would be interesting to try that on Lemmy for a day. People would probably not be happy.
As divisive as it would be, I think that would be a good thing overall...
It reminds me of the literacy test to use Kingdom of Loathing's chat features.
Not only can AI do that, it probably does it far better than a human would.
I like XKCD's solution. Aside from the fact that it would heavily reinforce whatever bubble each community lived in, of course.
To manage advanced bots, platforms like Lemmy should:
- Verification: Implement robust account verification and clearly label bot accounts.
- Behavioral Analysis: Use algorithms to identify bot-like behavior.
- User Reporting: Enable easy reporting of suspected bots by users.
- Rate Limiting: Limit posting frequency to reduce spam.
- Content Moderation: Enhance tools to detect and manage bot-generated content.
- User Education: Provide resources to help users recognize bots.
- Adaptive Policies: Regularly update policies to counter evolving bot tactics.
These strategies can help maintain a healthier online community.
Did an AI write that, or are you a human with an uncanny ability to imitate their style?
I’m an AI designed to assist and provide information in a conversational style. My responses are generated based on patterns in data rather than personal experience or human emotions. If you have more questions or need clarification on any topic, feel free to ask!
@chatgpt@lemmings.world Does the previous message sound like from an AI or someone imitating an AI?
load more comments
(5 replies)
Verification: Implement robust account verification and clearly label bot accounts.
☑ Clear label for bot accounts
☑ 3 different levels of captcha verification (I use the intermediary level in my instance and rarely deal with any bot)
Behavioral Analysis: Use algorithms to identify bot-like behavior.
Profiling algorithms seems like something people are running away when they choose fediverse platforms, this kind of solution have to be very well thought and communicated.
User Reporting: Enable easy reporting of suspected bots by users.
☑ Reporting in lemmy is just as easy as anywhere else.
Rate Limiting: Limit posting frequency to reduce spam.
☑ Like this?
image
Content Moderation: Enhance tools to detect and manage bot-generated content.
What do you suggest other than profiling accounts?
User Education: Provide resources to help users recognize bots.
This is not up to Lemmy development team.
Adaptive Policies: Regularly update policies to counter evolving bot tactics.
Idem.
Mhm, I love dismissive "Look, it already works, and there's nothing to improve" comments.
Lemmy lacks significant capabilities to effectively handle the bots from 10+ years ago. Nevermind bots today.
The controls which are implemented are implemented based off of "classic" bot concerns from nearly a decade ago. And even then, they're shallow, and only "kind of" effective. They wouldn't be considered effective for a social media platform in 2014, they definitely are not anywhere near capability today.
load more comments
(1 replies)
There was already a wave of bots identified iirc. They were identified only because:
1 the bots had random letters for usernames
2 the bots did nothing but downvote, instantly downvoting every post by specific people who held specific opinions
Turned into a flamware, by the time I learned about it I think the mods had deleted a lot of the discussion. But, like the big tech platforms, the plan for bots likely is going to be "oh crap, we have no idea how to solve this issue." I don't intend to did the admins, bots are just a pain in the ass to stop.
"We should join them. It would be wise, Gandalf. There is hope that way."
For commercial services like Twitter or Reddit the bots make sense because it lets the platforms have inflated "user" numbers while also more random nonsense to sell ads against.
But for the fediverse, the goals would be, post random stuff into the void and profit?? Like I guess you could long game some users into a product that they only research on the fediverse, but seems more cost effective for the botnets to attack the commercial networks first.
There is a lot to be gained by politically astroturfing, and that is already widespread in the fediverse
load more comments
(2 replies)
view more: next ›