Back in 2014, moot banned gamergate discussion from 4chan. This led to an exodus of the most adventurous and curious channers to 8chan, which had a strict free speech policy. For a few months at least, 8chan was a haven of smart free discussion on all sorts of interesting matters. For once, a chan was not a cesspool of shills, idiots, and school shooters, but an anonymous gentlemen's republic of letters. I exaggerate somewhat, but only along the true vector. Of course 8chan's lack of moderation led to it later becoming even more of a cesspool than 4chan. Curation matters a lot.
Anonymity evades enforced groupthink, but relies harder on good curation. Curation is political and taste-intensive, but laborious enough that it must be parallelized to many volunteers. This creates endless drama with "jannies", "trust and safety", shills, spammers, etc, who represent neither the poster community nor the administrator.
The solution is to empower the poster community to moderate as part of normal use of the platform, aggregating their implied and explicit judgements (eg votes, hides, interactions) as signals of quality. This requires much investment in a robust algorithm to enfranchise a high quality self-endorsing poster "eigencommunity" consistent with the vision of the administrator and to avoid empowering the herdthink of tasteless spectators. This can use many signals (eg posting history, offplatform clout, even demographic info) to curate posters. We can discuss details another time.
Much of this is also in asking the right questions of the herd. For example "hide" is unnatural for mere disagreement and thus a better signal of undesirable posts than "downvote". Balanced with "like", the overall value signal becomes positive-biased with no heckler's veto. Relatively objective consensus reality questions like "which board does/doesn't this belong in" are entirely harmless from a groupthink perspective. Voting systems only suck if they are badly designed.
Anonymous forums also rely more on topical boards to subdivide discussion, because you can't follow individual anons. You could also use a personalized feed based on interactions, but that makes shared perception of reality impossible. Feeds are bad UI because their state space is large, opaque, and even adversarial. Boards create a public map of where to find what, which can then be navigated and discussed reliably.
Nonoverlapping boards make discovery and cross-pollination hard, and make it hard to improve the taxonomy. Reddit, 4chan, and 8chan all suffer from this. The solution is overlapping boards ("tags"), with a single thread in multiple tags. Then a smaller tag focusing on some niche where it overlaps with a larger one isn't a problem, it's a solution: the smaller tag "indexes" the larger one. This means the entire site can be like one big board and everything can be well-indexed right from the front page by a loose hierarchy of tags. Meanwhile individual tags can still be well curated with respect to their individual topics and culture, and are much easier to start as new threads can be cross-posted for promotion and old threads can be retroactively tagged in.
This puts further weight on curation: tags themselves need to be competitively curated, and threads on the larger board pages will have to be filtered by value if they are not just going to be an unusable firehose. This is where the earlier investment in curation algorithms pays off again: board front pages can display some mix of recent and best, with further refinement available by drilling down into smaller boards. Meanwhile the same technology that aggregates crowdsourced judgement for questions like "is this spam" can be used for "what tags does this belong in".
The result is an anonymous imageboard that is as scalable as mainstream social media while being better indexed and even better curated. I think it has a lot of potential, so this is what I'm building.
> a high quality self-endorsing poster "eigencommunity"
Experimentally: you could make this a literal "eigencommunity" using PageRank, which calculates eigenvalues of a big recommendation matrix.
The naive way to score social posts is raw engagement, such as X likes. At scale this ranks by lowest-denominator appeal and rewards clickbait slop.
PageRank instead propagates reputation. A high-rank user's endorsement counts more. Quality smaller accounts get their own recommendations buffed transitively if big users endorse them. Conversely, if a spammy user has been downranked by many high-ranking users, then that user's own endorsements might even count negatively!
It's an elegant algorithm with an interesting twist: you can change the initial condition. So you could, for example, show PageRank where (admin) is defined as Score 1 and everything else flows from there. Or you could define *yourself* as Score 1, calculating a separate ranking for each login--in that world, if you are a spammer or botlike ideologue, you get hellbanned as a natural corollary of the algorithm to seeing more of your own kind. Many roads to Rome.
>>2855 Yes the current implementation of the “eigencommunity” of good posters is closely related to pagerank: your authority as a voter is a function of how much value (positive and negative) you have produced, with value being the revealed opinion (votes and replies, like links) of the high authority posters. There’s a nonlinearity in the function to spread the juice around to smaller anons and smooth out the extreme long tailed skew of the underlying value production distribution. Its not quite this simple, there are multiple ways of counting contribution and consensus alignment, there will be more features in future, and the hyperparameters will be learned to optimize which equilibrium we end up in, but you have the idea.
>>2857 >>2855 Also, i’m not going to hellban/heavenban everybody. Besides being rather more expensive (several orders of magnitude, naively) than computing a single canonical community perspective, the difference between different possible authorities is not as large as you might think (at least at this stage). There will only be a small range or a few viable equilibria, at least with anything like the system i have described or built. No point in these minor solipsisms.
Furthermore, this goes against the “shared public reality” principle that sofiechan is committed to: when you see the front page and the valuations there, that is the objective reality of the matter, and i see the same thing if i go to the same url. Thus it is possible to discuss and reason about the outputs of the algorithm, adjust for its biases in our own thinking, and generally treat it as “territory” and not get raped by it like some kind of cartesian demon. I’ll leave all that to twitter, instagram, and tiktok.
feeds are such a powerful concept that I wouldn't ditch them entirely. I agree you don't want individualized feeds, but it might be helpful to have separate feeds for each 'cluster' of users (like the old twitter sim-clusters). To get your desired transparency properties, you'd have to have a way for users to navigate between all the different cluster feeds.
Why not just use tags for this? The simcluster insight is that user engagement has structure that is not well-captured in language. Though ideally the community would then develop language to talk about this structure, which could result in a set of tags that more cleanly partitions the space of topics. But the answer should be to use all the signals you have and make them transparent and socially usable, rather than to forego the incredible power of content feeds.
>>2859 Well as you noticed yourself, tags are naturally closely related to “simcluster” feeds. Whats the difference between a named feed and a tag page? I think what you’re proposing with respect to the current tag scheme is that users should should perhaps themselves have tag connotations in some future version of the tag system that does unsupervised learning to auto-suggest new tags that exist in latent space but aren’t explicit yet. I had this planned using a carefully concocted sparse autoencoder of some kind over word vector thread embeddings, but it might make even more sense to run this over participant clusters, who are much more direct signals of social/interest clusters than even words. Thanks for the idea!
I think the tagging system is a good idea and there may be a scalable way to put the right tags on posts. The idea is to just leverage the high-dimensional vector embeddings for LLMs for the tags. So a post (or maybe a whole thread) can be encoded as a vector and each tag will also have a vector embedding based on some description. Then we can tag the post with the nearest tag vectors using some distance metric.
Definitely would need tuning to get right, but we know the high-dimensional embedding for LLMs is useful for encoding meaning in a loose way. Might be interesting to test out.
>>2862 Yes I've been thinking this. I don't think you even need an LLM. Good old fashioned bag-of-words document embeddings would probably work fine (and be lots more efficient). But also, the tags are going to be populated and maintained by human posters, and the effort of correct tagging is small relative to the effort of writing a thread, meaning we have a small amount of work to do relative to the amount of labor available, meaning fancy technology only gets us a small marginal benefit and probably isn't worth it. The main thing fancy technology could do which actually is difficult otherwise is to suggest new tags based on unnamed parts of the thread embedding latent space. But I think we would want human action to actually realize them, because again tags are actually social arrangements, not just an index. So really what we want is the minimum viable thread embedding latent space from which we can discover new tags (and we get auto-tagging for free even if it doesn't matter much). The other guy's idea of using the presence of particular posters for this is cool. But this is all far away. I'm going to finish the tag curation and creation system first, and we can re-assess when it becomes a problem (it may never become a problem).
Hey admin, this is Stooliths from X. You contacted me about your site in response to me giving a sort of slapsdash outline of my own personal idea for a forum, so in order to reply to your more detailed outline of your own site I will give a more detailed outline of my own idea.
A good place to begin is the place where my thoughts on this subject began. And that is with two easy to make observations: 1)4chan sucks. 2)Twitter sucks. I will explain why
Why 4chan sucks: I think it is fair to say that almost everybody who engages in "fringe discussions" online has used 4chan at some point. The reasons for which are obvious. 4chan, due to the fact every post appears in the catalog of the board anonymously in a non-hierarchical fashion, allows any given poster to receive as much feedback as any other given poster for a post. In theory this allows for any post to be awarded for its own merit. With interesting high-effort posts receiving lots of engagement, and terrible shitposts sinking to the bottom of the catalog. In practice this is not the case. Anybody who has used 4chan (or a similar site) for a long period of time will be familiar with a number of facts: The moderation staff behaves totally arbitrarily, the posts which receive the highest reply counts are almost always ridiculously low-effort bait posts, a good portions of threads on a board will almost always consists of scatological bullshit like discussions about transgender genital surgery or thinly-veiled interracial cuck porn, and if a poster if well known on a particular 4chan board it is because they are a mentally ill spammer. Now to a certain extent "low quality" threads serve a purpose, they are amusing (It's an off-topic thread, the best kind of /tv/ thread. Real discussion about television and films is for faggots. etc). The problem is, with an anonymous website that has no mechanisms for content curation, the low-effort people posting for shits and giggles and the idiots posting low iq hysterical bait will crowd out the real substantive discussion. And a lot of the people post for laughs aren't even funny anymore. So, in reality, 4chan sucks
Why Twitter (X) sucks: Twitter has a baseline level of appeal. You get to see what THE WORLD is saying about THE ISSUES OF THE DAY. Not only that, you get to follow specific users of your choice to see what THEY have to say. The best way to use Twitter was to browse the likes of posters you liked, it was automatic agent based content creation. Then they removed the ability to view likes and now you can only engage with posts through THE ALGORITHM and your feed. Now algorithms are aesthetically unappealing because they are not determined by a single human beings explicit choice (as great aesthetic things are). Even if you only browse the accounts of posters you like, they will inevitably replying to what they see on their feed and what they will see on their feed is more likely than not engagement bait. Engagement bait comes in many forms: reposting 8 year old /pol/ videos about jewish bolsheviks, reposting 8 month old Tiktok cringe by liberal women, pretending to be retarded and getting people made, saying the most clichéd median RW or LW "takes" possible in reply to some other high-engagement post. All of this is poisonous to intelligent discussion because people who wish to say intelligent things do so voluntarily not in response to what some faceless mass of people "like" or "quote". Now Twitter has the opposite problem of 4chan, there is only automatically curated content but the curation mechanism is actively perverse. So, in reality, Twitter sucks.
Now I have observed two problems. 1)Anonymity in an open board structure will lead to garbage content receiving the most engagement. 2)The "feed" as a curation mechanism leads to either hyper-agreeable or hyper-disagreeable content receiving the most attention. What is the solution to both of these problems? Voluntary user based curation
Allow me to describe how I believe this can be achieved. The general structure of the site would be like 4chan with multiple boards and a catalog containing a specific number of threads. You can open the threads and reply to them, and every thread starts with an image. This structure is good because it makes it feel as if you are in the public square, and it rewards good text and image based user contributions.
Here is my first difference with 4chan. In order to post you must make an account with a username. The accounts on this website will not be like the usual forum account because they will have a functionality outside of the public boards. Upon clicking on the username of a given poster you will be taken to their "private" page. The private page of a user will be much like the account of a Twitter user. Allowing them to share blog posts on a user wall.
The way I would tie the private and public sections of the site together is like this. A post on a public board may begin with a link to the private post of a user or it may not. Posts with a link to a private post will have a different outline on the catalog than other threads. In additions, private pages may choose to archive posts (from anyone) from public boards onto their own page. This is in addition to the ability to repost private posts from other users. The private posts of a user will have no reply section, in order to reply to a private post you must link to it on a thread in a public board. This makes it so that people will actually use both portions of the site in conjunction.
Upon following a user you will be notified upon them creating a private post, aswell as when they create a new public thread. If you like a user, you get to see what the user thought was interesting enough to post or share. If a poster likes you, they will get to see what you thought was interesting enough to like or share. This is what I call voluntary user based curation. Everything you see is based on your own free will choice. If you want to only see what a select few want you to see, you can. If you want to venture into the "public square", you can. Users with large followings (because of their good posts) will creates threads more people are notified of. This mechanism allows for high-quality posters to be rewarded while maintaining a free and non-hierarchical site design.
>>2873 that's a cool idea. You can post your blog on your own page, but all discussion is inherently public. With likes you're basically going back to the older twitter model of non-algorithmic retweet-like-curation, if I understand correctly. So if I understand, there's no structure to the public square except who has posted and like what and who has followed who? Everything is individually curated. What then is the difference between the user pages and the public square? Your model is twitter but pared down to a core non-algorithmic individual-curation model?
You should build it in any case. The world needs more niche social media, and people are getting tired of X becoming the big brother of 4chan.
>>2871 good summary of the situation. As I've said above what we're trying here with sofiechan is to fix the problems you identify with 4chan. >The moderation staff behaves totally arbitrarily, the posts which receive the highest reply counts are almost always ridiculously low-effort bait posts, a good portions of threads on a board will almost always consists of scatological bullshit Basically, sofiechan is targeting this problem, making it much easier for the highbrow core poster community to somewhat consistently enforce decorum on everyone else, and to emphasize the best discussions (by their reckoning) in the catalog. >"low quality" threads serve a purpose, they are amusing good "shitposting" is definitely an art form, but it might be an inherently lowbrow art form. Nobody has dared to attempt much of it around here yet. I wonder if we will ever develop that kind of culture, or if its basically impossible when the average post isn't shit.
>>2876 >With likes you're basically going back to the older twitter model of non-algorithmic retweet-like-curation, if I understand correctly. So if I understand, there's no structure to the public square except who has posted and like what and who has followed who? Everything is individually curated. What then is the difference between the user pages and the public square? 1. I wouldn't have likes. I consider them to be a superfluous feature if there is no algorithm 2. There are a few key differences between the public boards and a private page. As I envisage it one could gone a given board like "politics" and see a specified number of threads in the catalog. The threads are temporary like 4chan. If one clicks on the catalog he sees every thread that is not yet archived. The private pages contain things that are permanent. You can post something yourself, repost somebody else's private post, or archive a public post. Any thread can link to any private post, and any private post can link to any thread or private post (to allow for thought-out "Responses" over a long period of time. Think a formal "written debate"). So not everything on the site is necessarily curated, but you can choose to curate on your own page and you can follow what other people have curated. What it would look like in practice is the boards look like 4chan boards and the user pages look like Twitter pages with tweets and retweets.
Now there a certain trade offs in any social media structure. You can't have an anonymous community while also having a community which encourages individuality. Similarly you must trade off some ability to manually curate content for good automatic content curation. I think our two solutions are attempting to address different issues. You want to create an anonymous platform like 4chan which has a mechanism that prevents extremely low quality content from crowding the site. I want to create a community of individuals who are able to freely able to access the content of other individuals while creating a selection of preferred content. Basically I imagine old style "Twitter Spheres" interfacing with the same public forum.
One thing I want to specify about my idea in regards to shitposting is that I think it allows for shitposty replies while discouraging shitposting OPS because of the sharing mechanism. A high quality OP is very likely to be shared by many highly followed users (known for good curation). Once a thread has high engagement it is likely to have numerous replies of varrying quality. Think about a "tweet" on Twitter where a good portion of its replies wont be of a similar effort to what they are replying to.
>>2873 This is very close to our vision for our next social media project at FUTO. Some notes:
> The general structure of the site would be like 4chan with multiple boards and a catalog containing a specific number of threads.
Having a set number of threads is the wrong model. Threads should be able to last decades, as threads enable a specific mini-community with-in themselves. The 4chan solution of always reposting the same thread is much worse than the world4ch solution of just never deleting threads.
> Upon clicking on the username of a given poster you will be taken to their "private" page. The private page of a user will be much like the account of a Twitter user. Allowing them to share blog posts on a user wall.
In addition to this we're implementing a claim and vouch system, such that you can identify characteristics of a poster based on what others vouch about.
> The private posts of a user will have no reply section, in order to reply to a private post you must link to it on a thread in a public board. This makes it so that people will actually use both portions of the site in conjunction.
This is interesting but feels like an unnecessary restriction. You're likely going to get a higher quality discussion if you can enact some walls here, like private posts are limited to followers of a user (or the poster has the option to enable this).
Additionally, the forum format for posts (avatars, sigs, proper quoting, bbcode) is superior to the stripped down 4chan post. We'll be basing the look and feel off phpBB. Otherwise we're looking for very similar properties in a social media product, and have the more Twitter-like functionality already present in Polycentric.
>>2878 >>2879 Very cool, guys. Let a thousand alt social media flowers bloom. I look forward to seeing your respective attempts. I hope you’ll still want to come visit us here.
