There are in politics conversations, but still not nearly as bad as reddit. Even before I left, it was like a weird kind of prolific dead. Kind of like the conversations in OP's pics.

I appreciate all of the extra work you do in terms of Threadiverse infrastructure and quality of life.

Many Reddit bots have also straight copy+pasted content from Reddit or other social media with only trivial changes to the text or image, if any change, so the Threadiverse needs to be able to catch those as well. A better internal search engine, especially one that can search for strings of text [edit: and one which can search through deleted and removed content], would help users track down if an account’s content was routinely copy+pasted. I think a new instance (unaffiliated with any particular instance) staffed by users familiar with bot detection to flag bot accounts for federated instances to then ban would be the best facsimile of Reddit’s now defunct BotDefense subreddit, which was a critical tool for users to tackle the site’s bot problem.

This account I noticed yesterday is an example of a Threadiverse account just copy+pasting content (or in this case, crossposting to the original community) with little to no change. I have reported it to its host instance as suspicious but it has yet to be removed. An independent and informed instance for flagging bot accounts could more effectively communicate to the host instance as well as to Federated instances that this account is ticking the boxes of a bot account and should be blocked, banned, or at the very least closely monitored.

A detector for bot networks, such as in the screenshot above, would also be helpful. Some sort of indicator of if several accounts are interacting with each other or on the same posts as each other far more often than they are interacting with other accounts and other posts would be helpful.

Maybe like the New Account Highlightenator on the Voyager app, there can be an indicator for when an account has fewer than X amount of posts or comments (i.e. a potential new bot account), as well as an indicator of if the account has returned from a long hiatus of posting/commenting (i.e. a potential former human account that was bought or hacked to become a bot account).

I’ll try to think of more signs of bots and more ways the Threadiverse can build infrastructure against them.

Will this LLM detection be something my LLM prompt can include?

What angle are you approaching it from?

I think LLM detection is gonna be tough, I think a verification route and cutting bots off from the source is going to be ideal but not sure how to tackle that while respecting privacy.

I did have an idea of a app using nearby devices and after flagging 7 unique devices it's tagged as verified, interesting but not sure it'll work in rural areas.

Participate in closed platforms, where you can be reassured you’re engaging with real humans - but you’re also under total surveillance.

Closed platforms where the only propaganda bots are the ones controlled by the platform. They can then remove ADs from the business model and instead finance the platform by selling access to the bot accounts. And people will think they are in the perfect social media without advertising and only RealPeople™ that they can completely trust.

Except that Meta has already admitted that it is using AI bot accounts to "drive engagement."

After an outcry from real users, Meta said it had removed some AI bot accounts. But nothing else they said indicated that the experiment is over.

Eventually, social media is going to be nothing but company-generated AI bots, "bot farms" run by humans in developing countries, and (hopefully) a small number of actual users who can't tell the difference between those things and real people.

Well there's holes in the fact that resale of accounts is an active and common phenomenon, and creating a fraudulent identity for an online service (even if you have to doctor an ID template) is a low-risk barrier of entry.

Remember how people used death stranding photos to get around face ID? It's the same concept.

I'm not giving my fucking id to anyone.

There is no service I need to use badly enough to do that.

I won't use anything meta makes. Don't use Snapchat. Reddit is fucking dead as far as I'm concerned. I'm not on twitter, blue sky.

I know im not fully anonymous and there is shit I do that makes tracking me possible or even easy, but I'm not going to make it easy on anyone.

Go fuck yorself with your surveillance shit.

3. Closed systems with vettings and circles of trust. I don't need to know your identity. I just need to know that the person I know and trust knows and vouches for you.

Wow. I didn't even think of that.
Bots, of course, will be exempted from verification. See also: dating sites & apps. Unless governments (which ones?) handle and enforce things. But will they have the resources?

from a different account since feddit.online wont show me this comment

There's nothing new happening on this platform, literally the frontpage consists of:

• doomerism
• trump and friends news
• tech company does stupid thing
• political meme
• lets revolt guys. it'll happen any moment now we swear.

The majority of posts are made by the same people.

As someone here since ~2022, I've seen the frequency of posts continue to fall.

Totally feel this. It’s kinda wild how refreshing it is to see someone actually dig in instead of just shouting hot takes into the void. When people put in real effort — like, actual research, receipts, context, the whole thing — it reminds you that not everyone is just doomscrolling and giving up.

Honestly, posts like this are the rare moments where you remember, “Oh right, humans can be competent and thoughtful.” Gives me a tiny spark of optimism I didn’t expect today.

More of this energy, please.

thank goodness you put that /j there, i was close to issuing a downvote!

Dead Internet Reality

In Science, a theory already has proof. We just use it to mean hypothesis colloquially

When we post via snailmail ;)

It's a dumpster fire. Subreddit sim was at least trained on the individual subs so each bot had a different vernacular. Asslips is just someone mashing space on their predictive keyboard over and over and it frequently gets stuck in loops saying something is or is not trans.

What's the Lemmy community name?

completely unrelated aside...

i feel like we're overdue for another zombie-genre revival

You're totally right! The eloquence and creativity showcased in AI comments underscores how they cannot be distinguished from comments by human users.

most macbooks and android devices, too

Don't they have an apostrophe somewhere?

According to Wikipedia their apostrophe should be located where the US-ANSI dash - is located.

Huh, is this why I've been accused of being a bot before? I've been using the noun-verb username format since the late 1990s.

Yep I tracked those accounts for a while when I was still on reddit.

Blossombreezefairy, bubblypeachbliss, sparklepeachgiggle, tulipheartblossom, etc were all real bot accounts I'd found prowling around.

My theory for all these flowery names was that they'd be all assumed by unaware users to be women behind the usernames, and female-presenting accts would net a little more engagement and most importantly, karma, which is what builds the monetary value of these things.

From what I recall, the most common Reddit bot nomenclature was [word][word][number], [first name][last name] (usually feminine to later become a NSFW spammer), and [string of random letters and numbers of various lengths]. Each may have had hyphens, underscores, random typos, or deleted or duplicated letters. Of course, not every bot fit these archetypes.

[word][word][number] was also the default nomenclature for genuine human accounts created through a certain avenue, if I recall. If you go bothunting, be mindful of false positives triggered by one or two red flags.

Adjective-Noun-#

Fuck, I'm the inverse. I'll know better for next time.

There's also the phrase "It's not just X -- it's Y" that LLMs seem to looooove to use.

I've also found that a lot of them end their statements with a call to action, example: "Let's all strive to improve someone's day" or something like that.

That's a good way to describe it, too. Yeah. Amy Shira Teitel had this video the other day about how reading skills are dropping massively and how it links into short-form video and short-form entertainment in general.
And it made me realize: I like cozy reading every so often. Sure. Currently reading the third Legends&Lattes book, and it's fantastic as are the first two. But I could not do that all the time. It's too lightweight in prose and content for that.

Similar thing with AI text, and in particular in long-form AI-generated text. It's not text meant to keep you engaged while reading it, just not actively disengage you.

Yeah, LLM would probably not omit the final punctuation unless specifically prompted to or unless it is given a ton of examples of comments it should mimic in the prompt.

in fact you did in your comment

Whoosh

I agree. I believe it's difficult for me—or anyone else—to pinpoint a specific comment as being generated by an LLM. However, when numerous bots are involved, the pattern becomes clear. Essentially, many responses end up repeating the same points, just phrased differently and using the same keywords. If you were to use ChatGPT to summarize each response, you'd essentially get a very similar outcome from all the bot-generated replies.

Vibe coding.

List ends are exclusive, a new programmer could easily make that mistake.

Why would they cleaning characters of the end in the first place?

linkedinlunatics is a phrase for a reason.. LLMs are just polite white people faking the interaction. anyone who's worked in or around corporate recognizes the synergy

https://www.youtube.com/watch?v=GyV_UG60dD4