my leading contender for great filter.
posted by j_curiouser at 1:09 AM on August 14 [17 favorites]

What I would say to that is that we are at a crisis point with the rise of LLMs and AI. I have a hard time seeing the contemporary model of social media continuing to exist under the weight of LLMs and their capacity to mass-produce false information or information that optimizes these social network dynamics. We already see a lot of actors¡ªbased on this monetization of platforms like X¡ªthat are using AI to produce content that just seeks to maximize attention. So misinformation, often highly polarized information¡ªas AI models become more powerful, that content is going to take over. I have a hard time seeing the conventional social media models surviving that.

Don't threaten me with a good time.

I still have hope in my heart, against all objective reality, that social media could be an ongoing net positive to humanity. Having the infinite-growth-and-profit-no-matter-what-the-consequences architects behind both LLMs and social media destroy each other might be a healthy step in the right direction.
posted by slimepuppy at 2:32 AM on August 14 [5 favorites]

slimepuppy: If you don't know who Ed Zitron is, you'll REALLY like Ed Zitron.
posted by BiggerJ at 2:53 AM on August 14 [14 favorites]

The "basic structure" of social media is talking at people through small boxes, rooted in parasociality. Language, though, grew out of real world proximity¡ªour social fabric woven from person-to-person conversation and networks. Depriving communication of nuance, information of context*, and relationships of accountability¡ªoptimizing for brevity and rapid emotional response¡ªwas bound to subvert complexity evolved elsewhere over eons. And because the tool of language is effective whether or not we appreciate how it works, the degradation is largely lost on us.

*The web's first and best trick was the link. Therein laid the promise of depth and richness, of understanding both eccentric and holistic. Intent was essential, and exploration was its own reward.

Alas, we've found ourselves ensnared by the feed, the comment, and "content"¡ªthe unfurling of hyper-consciousness supplanted by corporate algorithms and people talking past each other.

As politics spiral downward and "media" is lost, the bulwark, as ever, is proximity, and patience.
posted by angelplasma at 3:40 AM on August 14 [24 favorites]

What worked for me on youtube was to kill google data collection completely, so that I don't get a feed and watching videos only from my subscription feed.

What we need I feel are better creator discovery services.

Or maybe let dislikes or other signals drive a creator rating which can then bubble up videos based on that.
posted by mahadevan at 4:09 AM on August 14 [3 favorites]

the way i see it, ai and social media have always been the obvious endpoints for the internet as it was designed: an unregulated international communications system with no public infrastructure. anyone lamenting the "promise" of computing is incredibly naive, some kind of huckster or both.
posted by AlbertCalavicci at 4:27 AM on August 14 [1 favorite]

What worked for me on youtube was to kill google data collection completely, so that I don't get a feed and watching videos only from my subscription feed.

This absolutely works on Youtube, but this method doesn't work on other platforms.

I have a a Facebook account with zero friends, have never liked a single post, posted anything outside of Marketplace, or even scrolled the main feed. However, the visible section of the main feed is filled with white-right-rage and seems to get more angry the less I interact with it.
posted by dobbs at 4:51 AM on August 14 [12 favorites]

I tried Ed Zitron's Better Offline podcast and he's not much different from Fox News or Rachel Maddow in the sense that it's just angry manboy nonsense pointing north instead of south. The man comes across as a huckster who knows Angry Sells and is trying to soak up the listeners too stupid to realize they're doing the same thing as those they're criticizing. As Hearst says in Deadwood: "I don't care what you write in your paper. I'll start my own and lie the other way."
posted by dobbs at 4:56 AM on August 14 [6 favorites]

i mean, i wouldn¡¯t call a show anchored by a queer woman ¡°manboy nonsense,¡± seems a little weird, even if it¡¯s a bad show.
posted by sickos haha yes dot jpg at 5:02 AM on August 14 [9 favorites]

This is a fascinating study idea that's worth pursuing, but speaking as a scientist who researches this area, this ArsTechnica article is far over-claiming what the first draft, which hasn't been peer reviewed, can actually claim.

Most importantly, it's not peer reviewed (yet?) so it hasn't been checked by other scientists. Second, it uses LLMs for simulations, which is a very interesting idea for research that hasn't been reliably validated in the world (this article in Nature summarizes some of the problems). Third, it tests design-based fixes for what are social problems. Imagine if someone set out to simulate whether a chronological feed would end gender based violence online. Yes, it's possible to reduce harassment somewhat with design changes, but misogyny is deeper than design. And since this paper is testing a design hypothesis, it's not surprising that the outcomes are "modest."

More fundamentally the Ars Technica article (and here I'm talking mostly about the reporting, though the authors lean into this a bit too) takes basic structures of human society and names them as problems that doom us if we can't "fix" them. At no time, for example, has there been a human society where the loudest and most influential people don't have outsized voice. That's a feature of all human communication in any setting, whether you're having a team meeting or talking online. It takes extraordinary measures (like Quaker meetings or Roberts Rules of Order) to constrain those patterns, and I'm not sure many people want to live in a world where all conversation is constrained to Robert's Rules.

I think it's a very clever project, worth doing, and will probably contribute substantially to a conversation about whether and how to use models like this in social and computational research. But it would be nice if Ars Technica writers believed enough in science to trust the peer review process.
posted by honest knave at 5:33 AM on August 14 [73 favorites]

Flagged as fantastic, honest knave.
posted by kristi at 6:05 AM on August 14 [3 favorites]

honest knave, I had similar questions about how LLMs were used in the study. From the OP:

Our idea was to start with the minimal bare-bones model and then add things to try to see if we could reproduce these problematic consequences. But to our surprise, we actually didn't have to add anything because these problematic consequences just came out of the bare bones model. This went against our expectations and also what I think the literature would say.

If they used LLMs trained on existing social media, wouldn't you expect their output to replicate existing social media?

This may be addressed in the paper; I only read the article. I was waiting for the interviewer to raise that question, but they didn't.

There's some interesting discussion of social media dynamics in the OP, but IMO the way the study is described seems to beg the question.
posted by postcommunism at 6:36 AM on August 14 [5 favorites]

dobbs: the visible section of the main feed is filled with white-right-rage

TikTok is like that for me - when you sign up for TikTok all it asks for is age and an email address, and I've tried five different email addresses (three of which were 'sockpuppets', no connection to my personal self), different ages ranging from 20s - 50s, and with a completely clean account (deleting app and data from phone each time)-- if I scroll just a little bit, without interacting, it become slightly-pro-Trump, a little more it gets MAGA-crazy, and if I keep going it goes into flat-earth-antivax-deepstate-epstein-conspiracy bullshit. I can't be any less identifiable and it goes down that path.

Instagram is relatively good about letting me scroll without showing me things I didn't ask for (every 30 days I have to turn off suggested things in my main feed, but that's easy to do); and, really, a lot of the Instagram content I follow is literally cool videos stolen off TikTok, so maybe that's my algorithm right there.
posted by AzraelBrown at 6:36 AM on August 14 [5 favorites]

okay so i am compelled to give my hot take hypothesis which is that pretty much all of what¡¯s wrong with social media is also wrong with the searchable web.

if we want an Internet that¡¯s not awful it¡¯s got to be an Internet where you can only find things by following links from people you already know rather than by using a search engine.
posted by bombastic lowercase boatshoes at 7:29 AM on August 14 [4 favorites]

they're doing the same thing as those they're criticizing

dang. I haven't seen a proper both sides in ages.

I find Zitron's newsletter a fine source of information. Sometimes Zitron provides some good criticism, and sometimes I enjoy a rant, and sometimes Zitron sounds a bit exhausted, a bit exasperated, and sometimes a big mad. It's 2025. I feel a bit emotional about everything going on, I certainly don't expect him to be perfect.

So maybe his podcast is too sensational - where is he lying? Where is Maddow confabulating, where is her many lies, like Fox News shovels bullshit out every minute? Because the right wing media is a garbage factory that creates a whole fantasy world of white male grievances.

Maddow and Zitron are not angry manboy nonsense. That claim is categorically false.
posted by zenon at 7:35 AM on August 14 [12 favorites]

the mechanisms that give rise to the worst aspects of social media: the partisan echo chambers, the concentration of influence among a small group of elite users (attention inequality), and the amplification of the most extreme divisive voices

All of these mechanisms can be at least partially mitigated by single-threading, go team metafilter. There are power users here, but no one gets promoted to hyper-publicity by an algorithm.
posted by justsomebodythatyouusedtoknow at 7:36 AM on August 14 [4 favorites]

Maddow and Zitron are not angry manboy nonsense. That claim is categorically false.

Both choose to enrage rather than inform, just like Fox. Both cover topics most intelligent people are already aware of but provide a "you're not gonna BELIEVE this" angle. You don't leave their shows smarter; you leave their shows angrier. That's my very definition of angry manboy nonsense.

I haven't seen a proper both sides in ages.

Yeah, I didn't do this. I picked two specific people. There are plenty of responsible people on the left who cover news by providing information in an objective and intelligent way, leaving you with the goods to make up your own mind.
posted by dobbs at 8:24 AM on August 14 [4 favorites]

dobbs, are you arguing that it doesn't matter whether the core claims are true or false, because ANY show with a, let's call it, "linkbait", style leaves its viewers dumber?
posted by PresidentOfDinosaurs at 8:52 AM on August 14 [7 favorites]

dobbs: "I have a a Facebook account with zero friends, have never liked a single post, posted anything outside of Marketplace, or even scrolled the main feed. However, the visible section of the main feed is filled with white-right-rage and seems to get more angry the less I interact with it."

Hm, weird. I guess MAGA is the default shit it posts. My FB is a happy place filled with rainbows, crafts, and cute animals and maybe once every 3-4 months FB tries to float a turd into my pool, I nuke it from orbit, and then all is well.
posted by jenfullmoon at 8:54 AM on August 14

I am probably 95% overlap with Zitron's outlook but after that AI piece that did the rounds recently claimed 35% of the economy is based on the AI bubble and will vanish when it does I realised his analysis was financially illiterate (the tech giants have other sources of value, they won't go to zero, stock markets are not the economy, etc etc). And I'm far less inclined to read him as a result.

As to the paper, it doesn't reflect my last year or so in a backwater of the Fediverse. Assuming there is something in it, perhaps the active moderation and the genuine federation of many servers has something to do with it, since the simulation seems to have been of a single unmoderated service.
posted by i_am_joe's_spleen at 9:30 AM on August 14 [6 favorites]

I have some questions about the methodology. Since most LLM models were trained in part on existing social media (e.g. Reddit), wouldn't the LLM agents used for this study be predisposed to reproduce existing social media dynamics, regardless of what tweaks were introduced to the system by the researchers? I'd be happy to hear otherwise if someone has more knowledge of the matter.
posted by ourobouros at 9:32 AM on August 14 [3 favorites]

Both choose to enrage rather than inform, just like Fox. Both cover topics most intelligent people are already aware of but provide a "you're not gonna BELIEVE this" angle. You don't leave their shows smarter; you leave their shows angrier. That's my very definition of angry manboy nonsense.

And as has been pointed out, that's actually a bad definition, based on the idea that anger is somehow bad in of itself. The problem with Fox isn't that it gets people angry, but that it gets people angry over lies.

It's okay to be angry about things like injustice! In fact, injustice should enrage you.

I am probably 95% overlap with Zitron's outlook but after that AI piece that did the rounds recently claimed 35% of the economy is based on the AI bubble and will vanish when it does I realised his analysis was financially illiterate (the tech giants have other sources of value, they won't go to zero, stock markets are not the economy, etc etc). And I'm far less inclined to read him as a result.

This, frankly, is the bigger problem with Zitron - he's too willing to let his anger lead him out over the edge into positions that he can't support, which in turn undercuts his argument as a whole.
posted by NoxAeternum at 9:54 AM on August 14 [7 favorites]

if [you] want an Internet that¡¯s not awful it¡¯s got to be

MetaFilter: find things by following links from people you already know!
posted by otherchaz at 10:00 AM on August 14 [3 favorites]

I actually think the main insight of the interview is that there is a power law effect that internet revolution causes--due to many-to-many network structures--which is fundamentally different from "actually, we had demagogic voices throughout human history". And once stated explicitly it is a nice intuitive insight, anyone could have thought of it.
posted by polymodus at 10:29 AM on August 14 [4 favorites]

This, frankly, is the bigger problem with Zitron

The problem with Zitron is that he is laser-focused on events and companies in Silicon Valley and perceives that as being the whole of the world. It¡¯s like someone writing about tech in the late 90s and never setting foot outside Redmond or engaging with any non-Microsoft software: there¡¯s a lot to get angry about and justifiably so, but it¡¯s neither the whole of the world nor what¡¯s actually interesting in that time period.

The story of ¡°AI¡± in 2025 has been consistent hit after hit of China¡¯s tech giants and startups roundly kicking Silicon Valley¡¯s ass on efficiency (DeepSeek, Qwen) and proving they can deliver equivalent or near-equivalent results at vastly reduced scale (Qwen, Kimi, Wan). The only major things worth talking about from the US side in the past 8 months are: Google broke their tradition of ¡°great research, shit results¡± and launched two genuinely excellent products (Gemini-2.5 Pro unquestionably dethroning Claude-3.7 as the code assistant of choice, and Genie 3 as 2025¡¯s only ¡°holy shit¡± moment so far), and after years of being the butt of jokes xAI finally put out a competitive model in the latest Nazi AI release.

Every other piece of news: DeepSeek dropping a zero off costs/carbon footprint and open sourcing everything important not just the weights, Alibaba releasing a sprawling family of models (Qwen 3) at every size range and consistently getting results you¡¯d expect from models 2 or 3 times larger, then releasing Qwen-Image (the first diffusion model with GPT4o-level prompt adherence) and open sourcing it, and now Wan-2.2 becoming the first real Veo competitor for video, also released as open source (and insanely high efficiency for video gen)¡­

2025 is the year of China¡¯s tech sector ending their second tier/copycat mentality and leaping ahead of Silicon Valley. And doing so in a way that brought the latter¡¯s moral failures sharply into focus. From carbon footprint to protectionist proprietary bullshit - the obsession with reaching ¡°first Trillionaire¡± status has unmasked the American tech sector as fully monstrous even as they lose their technical dominance.

And in Zitron¡¯s world none of this happened, or mattered, and I think he unwittingly does a lot of Silicon Valley¡¯s work for them by pretending like they¡¯re the only game in town, when aside from Genie 3 everything interesting is happening somewhere else.
posted by Ryvar at 10:38 AM on August 14 [12 favorites]

in Zitron¡¯s world none of this happened

maybe it happened a little
posted by mittens at 10:56 AM on August 14 [1 favorite]

earlier this year i made a finsta to separate my normal friends & family from my "going out raving" activities.

because i follow a lot of queer and trans djs, instagram very quickly figured out that in that i was very interested in a vector of "trans" content ¨C but not its direction. for a while my explore page consisted of instagram asking: do you LOVE trans people or do you HATE trans people?

and boy howdy there was a lot of hateful content.
posted by pmv at 11:43 AM on August 14 [3 favorites]

> metafilter: find things by following links from people you already know!

pretty much the only reason i'm here. well, that and seeing how far i can take experiments in style without becoming actually incomprehensible.
posted by bombastic lowercase boatshoes at 12:33 PM on August 14 [3 favorites]

seeing how far i can take experiments in style without becoming actually incomprehensible.

lol, we know, my guy
posted by sickos haha yes dot jpg at 12:45 PM on August 14 [5 favorites]

in retrospect i should have made my last comment one sentence without any punctuation but i missed the edit window so please in your mind edit it to be like that thx
posted by bombastic lowercase boatshoes at 12:47 PM on August 14 [3 favorites]

I never thought of it in those terms but if MetaFilter is my algo I'm good with that

keep suggesting content
posted by Didymus at 1:32 PM on August 14 [3 favorites]

This hits hard, especially the point that opting out doesn¡¯t actually insulate you even ¡°legacy¡± institutions like The New York Times are reshaped by social media logic. The idea that clickbait dynamics are upstream from not just politics but culture itself feels more true every year.

I think the scariest part is how the platform incentives are invisible most of the time we absorb them through tone, pacing, what gets signal-boosted. And once that logic infects not just media but how we talk to each other, the whole ¡°just log off¡± argument feels kind of naive.
posted by FifthCupOfTea at 2:15 PM on August 14 [7 favorites]

a power law effect [¡­] anyone could have thought of it

Iirc Shalizi et alia did
posted by clew at 4:30 PM on August 14 [2 favorites]

ourobouros: " Since most LLM models were trained in part on existing social media (e.g. Reddit), wouldn't the LLM agents used for this study be predisposed to reproduce existing social media dynamics, regardless of what tweaks were introduced to the system by the researchers? "
I wondered the same thing - if there's a feedback loop built into their model that could only be removed by training an LLM entirely on content that isn't part of social media.
posted by dg at 6:03 PM on August 14 [1 favorite]

if we want an Internet that¡¯s not awful it¡¯s got to be an Internet where you can only find things by following links from people you already know rather than by using a search engine.

This neuters a lot of the power of the internet. Whether you are a trans youth in some biblical hell hole or just an American wanting to learn about fly fishing in Zimbabwe there are a lot of gaps in the knowledge of people we(I) already know. Even if you want to stretch the people you already know to something like AskMe, you need someone who knows about it to point you in this direction.
posted by Mitheral at 5:45 AM on August 15 [6 favorites]

The missing first step is "get to know the right people". Sometimes easy, sometimes not, sometimes dangerous.
posted by i_am_joe's_spleen at 8:04 AM on August 15 [3 favorites]

> this neuters a lot of the power of the Internet


hand-maintained curated web indexes.
posted by bombastic lowercase boatshoes at 1:16 PM on August 15

A lot of work. Yahoo abdicated that game long ago, and had decayed to demanding payment for inclusion before they called it quits. There are orgs out there that try that approach even today, Curlie is one of them. It seems pretty ramshackle though, and has significant points of potential failure. Maybe a Fediversal approach might work?
posted by JHarris at 2:16 PM on August 15 [1 favorite]

This is the most garbage study I¡¯ve seen in a long-ass time. Crawling of my hole to scream ¡°you can¡¯t use LLMs to model human behavioral experiments you fucking weirdos.¡± Christ alive. Even an undergraduate understanding of human psychology would have shown the authors why this is the least valid study design you could possibly concoct for the thing they¡¯re purporting to measure.
posted by brook horse at 4:05 PM on August 15 [5 favorites]

Yahoo abdicated that game long ago,

"I've seen things you people wouldn't believe. Newsgroups engulfed in flame wars off the Google Groups home page. I watched blink tags flash on wallpaper in the Geocities Webrings. All these moments lost in time, as expired domains."
posted by otherchaz at 5:13 PM on August 15 [5 favorites]

This is the most garbage study

Except one scientist's garbage is another's treasure. Part of their thesis is that even with a rudimentary and inherently biased model of agents, it is the many-to-many communication network structure that predicts the same trends that we see actually happen in online social media. The network effect overpowers our individual different qualities.

Also, such heated reactions also proves their point. We're all nice reasonable people IRL, but the internet is so structurally threatening that we have this compulsion to rhetorically destroy foreign or flawed ideas.
posted by polymodus at 3:10 PM on August 16

What?? "This paper might be extremely wrong, but your reaction to how wrong it is proves it's right?"
posted by JHarris at 5:04 PM on August 16 [4 favorites]

I mean if someone IRL came up to me and directly claimed what it says, I'd be mad about it. Because people don't usually say such directly stupid things in person.

But if I had read about it, or seen it on TV, and I had it somehow pointed out to me they were relying on experiments carried out through LLMs because (to maintain metaphorical parity) I don't see it in the text of the FPP here, I'd be pretty upset about the study. I'd probably rant aloud about it, and have a petty little stamp around my petty little circle, that no one would know about, because I wasn't on the internet.

I don't think being mad about LLMs being used as an experimental mechanism proves anything. I think that is pretty flimsy reasoning, not the least reason for being we're not even on social media right now.
posted by JHarris at 5:14 PM on August 16 [2 favorites]