This is actually my hope for AI-gen content as well. That after it gets so 'good' that people genuinely can't distinguish it from reality anymore that they'll retreat (or return triumphantly rather) to the physical world to gather truthful fulfilling experiences and dopamine.

For a while there were a lot of posts from people experimenting with ChatGPT to write anger bait posts on Reddit where they would later edit the post to say it was fake, written by ChatGPT.

I assume they thought they'd be teaching people a lesson by making them feel foolish for responding to AI stories, most of which were too fake to be believable.

However it did not matter. The posts remained popular and continued to bring in comments even after the admission that they were fake. In advice subreddits, commenters continue to give advice on the situation. Some comments would say they saw the notice that it was fake but continue arguing about it anyway.

This makes a feature of Reddit very clear: The truthiness of a post doesn't matter. The active commenter base on popular subreddits just wants something to discuss and, usually, be angry about.

In retrospect it's obvious given that misinfo posts were the easiest way to karma farm for years even before AI.

I think it's going to effectively kill public chat communities without either proof of identity or attestation through a web of trust. Or rather turn them into little better than comment sections on news sites; thriving but worthless.

I'm active in a number of online communities that are doing just fine but the difference is those all involve ongoing relationships, built over time and with engagement across multiple platforms. I've no doubt this clock is ticking too but it's still harder to fake a user across a mix of text chat, voice and video calls, playing an online game, etc and when much of the web of relationships extends back into real life activity.

But I agree the golden age of easy anonymous connections online has ended.

> Online communities are definitely dying. I guess I hope that maybe IRL communities have a resurgence in this wake.

Would be super fascinating to watch play out. I grew up before the internet so, historically, I know how to seek out external communities, but by early high school I was deeply entrenched in online life - so I'm very rusty with finding new IRL clubs, cliques, etc. Fortunately my life is full of many friends and I go out frequently, regardless. For those younger people that never had life without the internet, I wish them luck on their search but at the same time I'm very curious to witness their journey.

Reddit is more or less dead to me, as the popular subs are botfests and the niche subs are empty. I'm lucky to get a single reply on gaming subs.

Did you ever introspect about who ruined Reddit?

IRL communities have to have some guides because a lot of people forgot how to gather. It can be seen among kids - try to give them soccer ball and see what they do with it :)

Yesterday I was watching people on the street and on the tram. Every other person was staring at their phone and scrolling through something.

That might scare me more than the fact that someone is chatting with an LLM bot online.

(I am pro-ai, use it every day for coding that I couldn’t achieve pre-2022 as I am lame coder.)

Serious question: If there are so many LLMs on online forums, who is doing it? Is it just 1000s of research students or something more nefarious? Is it AI businesses building up evidence that their output is as highly scored as humans therefore "buy our software"?

Reddit sold it's data to AI companies for training[1]. They could have refused, but companies like OpenAI likely would have harvested that data anyways. As such, it should not be surprising that AI models are pretty good at generating reddit posts. They were specifically trained to do that.

This is sad, because Reddit remained one of the final bastions of human content on the internet. For several years, appending "site:reddit.com" to a google search was a valid way to get something usable out of a google search. Doing that is still an improvement over raw-dogging Google's ranking algorithms with an unfettered search, but AI slop increasingly is the result.

This is one of my great disappointments in the current rise of AI. LLM's can give good search results when dealing with a topic they've been specifically trained on by human experts, but they're not good at separating human-produced signal from AI slop noise. We've done nothing to prevent a sea of AI slop from being dumped on top all the human signal that's out there. When AI companies enter their enshittification phase and stop investing in expert human trainers, the search results LLM's produce are going to fall off a cliff. Search is a bigger problem than ever.

_____

[1]https://9to5mac.com/2024/02/19/reddit-user-content-being-sol...

> I do know for a fact that many "users" here are LLMs.

HN autokills comments it detects as LLM. I think maybe you're not giving HN enough credit. :)

Public* online communities are dying. Discord is thriving

There's this old meme where someone asks what will happen when AI bots posts helpful, curious and thoughtful messages!? That's mission accomplish :D They can't be better then the average human though because of training data, so I don't worry about AI comments getting up-voted by real humans, I am however worried about fake upvotes.

Reddit users are definitely smartening up on many subreddits to the same kinds of engagement happening.

There was a post today that Google introduced unbreakable capture that required unrooted phone to pass its QR code.

We may end up with things like that…

> As I went through the posts it wrote I realized that as a reader I would have NO idea that these were just written by a computer.

I don't suppose you could show some examples? How convincing is the state of the art now?

Doesn't help there is that feature that hides the user's posts and comments

This kind of thing made me imagine the creation of "digital towns" the other day.

Imagine an online community where you can only join on the recommendation of two other members, who you must have actually met in person, to participate. Meanwhile, you leave at least some of the activity publicly available to the general public so that interested parties can meet up IRL and join.

This could probably be implemented easily on top of existing online platforms like Discord, Reddit, etc. since it's really just a community building rule, not a community itself.

Communities in FB, WhatsApp, Telegram etc are actually flourishing. As it appears real time gated communities are doing fine.

It’s an unpopular opinion but I am looking forward to ID and age verified social media. If done right we can have real people around again.

BTW, ironically the harsher communities like 4Chan doesn’t seem to suffer from the dead internet. I guess it’s either because the advertising value is too low to justify AI use there or maybe AI API providers refuse to work with such a content this reducing opportunities to infest with bots.

> Online communities are definitely dying. I guess I hope that maybe IRL communities have a resurgence in this wake.

You can have both IRL and online-free-of-bots. I already wrote about it but one of the very best forum I'm a member of, where real people are posting, requires to be vetted in, web-of-trust (but IRL) style. It's a forum about cars from one fancy brand and you can only ever join the forum by having a member (I think it may be two, don't remember) who's already in confirm that he saw you driving a car of that brand. It's not 100% foolproof (someone could be renting the car for two hours and show up at a cars&coffee or take a friend's car etc.) but this place really feels like a forum of yore.

And people do eventually travel, so it's bound to happen that an owner shall go to another country, meet someone there, vet him in etc.

Now, sure, it may not be the "1 million users acquired in three days thanks to my vibe-coded app" scenario but that is the point.

You can imagine other domains where IRL communities have local groups, but where forums regroup different IRL communities all interested by the same hobby/topic/domain. And when people travel and meet, the vetted members do grow and connect.

Oh and on the forums a lot of the posts are pictures, where "Julian xxx" met "Black yyy Cyril" and you see both cars (and from more than two people): suddenly it becomes much harder to fake a persona... You now need to fake both Julian xxx and Black yyy Cyril and fake the pics. And explain why your car has never been posted by any carspotter on autogespot etc.

You can imagine the same for, say, model trains: "Met Jean at the zzz meetup, where he brought his wonderful 4-8-8-4 'big boy' locomotive, I confirm he's into the hobby, vet him in".

Naysayers and depressive people are going say it cannot work but I'm literally on one such forum and it just works.

P.S: if I'm not mistaken in the past in some nobility circles you had to be vetted by up to sixteen (!) other people from the nobility that'd confirm they knew you, your parents, etc. before you'd even meet the king/emperor/monarch to make sure that someone from far away couldn't come to, say, Versailles or Schonnbrun pretending to be a baroness or count or whatever. Quite the extensive check if you ask me.

Unless their account is <1 year I wouldn't assume they are a bot.

> Online communities are definitely dying. I guess I hope that maybe IRL communities have a resurgence in this wake.

Good. Tech has been artificially propped up to generate liquidity in markets. Never had organic demand or traction. Most output was trite and meaningless; just capitalism.

Zero cares if all the HNers rich on tech stocks go broke.

Happy to project as little concern for their well being as this forum has projected at other's fields tech eliminated.

Some wild Dark Triad shit peddled here where there is no obligation to humanity's problems, just HNers own personal preferences.

Yeah ok that can be projected back, terminally online script kiddie; fuck your stock portfolio and I hope you end up in public housing assistance.

Reddit was already on its way way before this LLM craze, hopefully the recent tech-related changes will only accelerate that process.

More of a philosophical question but if you have no idea whether it's a human or robot, does it really matter? Personally I dislike AI slop only when I can tell it is...

I find it amusing that this is the top comment. Reddit is so awful you finally wrote it off, but not before you used it to try to “karma farm and do some covert advertising”. It’s on-brand for HN hypocritical bullshit. But, since we are slamming on Reddit anyways without realizing how fucked HN is by the same petard, have an upboat fellow traveler.

How do we know now that this comment wasn't written by LLM?

I feel you. Especially in the larger subreddita. i participate, and mod, a few small ones, and the community there is pretty strong and folks shut down ai slop pretty quickly.

I'm not saying being a mod means it's bullet proof, but i do notice smaller communities tend to self police better and know what's real.

That said, your experiment scares me as well.

> I do know for a fact that many "users" here are LLMs

What factual basis do you have for that?

It's easy to botspam Reddit because even the real users always acted like bots. The big subreddits were the worst, but contrary to how the users keep saying "it's good if you find the right subs," no it's not. Wrote that place off like 10 years ago.

It might come down to shareholder/IPO stuff but you can tell Reddit doesn't actually care to put the effort in to crack down on bots (however you'd do that) because they already don't give communities proper moderation tools/third party tools and the site does censor

Whatever allegiances (with people, or allegiances to ideas) Steve Huffman has, or people like him - it's not enough. It's a site seemingly killed by greed

(Yes, I know moderating this stuff at scale is hard)

- A human. Beep boop.

I wonder, how much of the discussions on the results of agentic coding is just LLM slop.

Do you have an example of comments people engaged with?

> where I had an agent karma

Was this a browser using agent? What did you use?

> I am not quite there with Hacker News but I do know for a fact that many here are LLM's.

Please don’t do this here.

Unless you've discovered the secret sauce, LLM comments are very obvious. Even Altman revealed that they focused on coding at the expense of writing.

On the other hand, I’ve been accused of being AI/bot and if I say things the mod doesn’t like and is not their favorite thing to hear I’m “flamebaiting” or engaging in personal attacks when pointing out specific things.

Frankly, online communities have been doing for many years now, when the censorship, anti-free-speech, tone policing mods and mobs started dominating online and America really did not have the self-respect or confidence anymore to enforce the Constitution online.

> I do know for a fact that many "users" here are LLMs.

Name and shame.

[dead]

[dead]

I've been on the Internet for decades at this point and one thing I've noticed is that communities that, for example, ban political topics actually mean "positions I don't like" as "political". This is somewhat related to the Overton window but really a bunch of (mostly conservative) ideas get normalized so aren't deemed "political".

I see the same thing with "AI Slop". Yes, there is AI Slop but (IME) it's pretty easy to spot. But what's more annoying is how often people are willing to throw that accusation whenever someone takes a position they don't like, much like the "political" label. It's lazy and honestly just as bad as the slop itself because it unintentionally launders the slop in a "boy who cried wolf" kind of way.

I also have a theory that some AI slop isn't inherently successful. It's just heavily botted by people who are interested in promoting certain positions. I bet you could make a pro-administration LLM bot and another one promoting a communist revolution and no amount of model tuning would make the second as popular as the first because the first would hit third-party botting as well as platform content biases (eg Twitter).

I've personally been accused of being a bot. This is particularly true in recent time as I've tried to share facts and fact-based analysis of, say, what's going on with crude oil markets, the military operation in the Gulf and the politics and economics around it. I even saw one hilarious comment saying (paraphrased) "the bots are getting clever and posting about unrelated topics". This was funny because it never occurred to this person that no, it was just a real person posting something you disagreed with.

> I am not quite there with Hacker News but I do know for a fact that many "users" here are LLMs.

This just makes me wonder...so what?

Some of the oldest posters here with the most karma continue to post absolute garbage takes on topics ranging from US healthcare to history of USSR, that are trivially disproven by learning the very basics from a Wiki article (e.g. not a high bar).

To be fair, this opinion slop is also present for new users and LLM bots, but is one kind really worse than the other, if both of them contribute to killing the community?

We already know what kills communities. It's the eternal Septembers. Infighting within leadership also doesn't help, but time and time again it's the influx of too many new users that nosedive and drown out quality contributions.

So you ran an "experiment" where you deliberately made someone else's community worse to see what would happen? Cool project.

Dead Internet theory ?