I feel the same way about cancel culture on twitter -- it is bad in the short term, but if it incentivizes people like Yann Lecunn to leave it -- it reduces the importance of the medium itself.
I hadn’t seen that. Thanks for pointing it out.
The thing is, I would love to help (and listen) to those who have studied this extensively. But there wasn't any information to go on. The article makes it seem like they shouted him down while delivering no actionable advice.
My understanding was that it was mostly coming from activists that don't really understand how AI/ML works.
Worse. It makes it seem like they shouted him down while not even delivering any concrete criticism. If that's accurate you can't blame LeCun for leaving. And looking at your typical political "discussion" on Twitter, I wouldn't be surprised if it is.
I pay $5/month to get Matt Taibbi’s emails, usually about two a month with in depth research and reporting.
In general I like to spend about five minutes a day quickly reviewing news on Apple News+. Otherwise I read good books, with a little time on HN. I use Twitter and FB for shamelessly plugging my books. Anyway, this works for me.
There's always going to be people who're interested in insightful, well researched articles and opinions. Isn't a part of the problem connecting these people, to the actual individuals doing posting in a more transparent way than a news-paper?
Having this as the default would pose problems of it's own, and make some of the deeper journalism requiring months/years to complete more difficult though.
It's also totally different than most journalists who push out stories left and right on topics they have layman understanding of. The quality and depth of work is significantly higher. I would pay for news reporting from gwern than random journalist.
Granted, Twitter is still a horrible place for long-form news (which is why it links to the full articles) and an even worse one for discussions. Which is why a different platform for instant updates could be beneficial, since news will have to adapt anyway.
If only there was a really simple way to syndicate all this content that all websites could follow and multiple clients could consume...
The reason is that publications no longer have the benefit of the doubt when they publish something wrong. If a story with a malicious source runs, that publication can't argue that they were duped by a bad source. The community will automatically accuse them of pushing fake news, regardless of whether or not it was intentional. The media no longer has any benefit of the doubt due to the intense political opposition.
Also contributing to this is a publication's desire to be first to run a breaking story. This creates a scenario where a publisher has to decide between waiting for additional sourcing to corroborate a big story or risking losing out to a competitor who may run the story first.
It's one of the main reasons that I try not to actively support Twitter-driven journalism and instead put my journalism dollars towards publications that do long-form writing that takes time and patience to produce. The late-breaking Twitter world is just too often a Kobayashi Maru situation for publications.
Which is why Twitter is probably the GoTo for news.
An article from vice on any company will mention at least three of their own previous posts and rehash company evil before getting to the point. They will add a paragraph full of questions with no answers and another one for speculations that are only somewhat related.
Why bother with that when you can get to the point using a small tweet?
There is something online news agencies are doing wrong otherwise you wouldn't see so many startups opening news sumarizers, curation etc.
Given that Twitter is the only platform where interesting ML discussions happen daily with a large amount of actual researchers and practitioners talking about their work directly I'm not sure in what way it's better to have people leave without an alternative. The value just decreases for everyone.
Not a community-specific service, precisely; I'm not talking about the sort of partitioning StackExchange does. I wouldn't want a CS-people-only Twitter. (Cross-pollination of ideas between fields is important!)
Nor am I talking about individual independent forums that have their own specific (rivalrous) memberships, where you can't expect network effects to guarantee that J Random Peer-in-your-domain will definitely already be a member. Everyone you wish would be there, should already be there. Like Twitter.
But different domains of discourse have different standards/norms, and it'd be helpful to keep each such domain separate, such that people don't try to interact using conflicting standards/norms; but rather will know what the primary set of norms are for their current domain, and will "mode-switch" to using those when they're in that domain.
For example, all of "academia" is one domain of discourse, with its own social norms.
So we could have one domain-specific Twitter-alike where, to join (and not get banned), you'd have to obey the rules of academic conduct; and to get verified (and so to make public tweets), you'd have to be actively in academia in some way (e.g. maybe you'd need to submit proof-of-identity in the form of the DOI of an academic paper you've written.)
Then you could have a separate domain for politicians and talking-heads, where the rules of discourse to follow are those of a political debate, and where user verification happens through a petition of non-verified users, proving that the user has "constituents."
I would imagine a blended app where one user account could be active on multiple domains at once, perhaps even with new posts from them collected onto a merged timeline for easy reading; but it'd be important to make clear, for each post, which domain it's in (e.g. by embedding the post within a background theme that represents that domain.)
I'd also imagine that getting verified in a second domain after already being verified as something else, would be harder, requiring a manual audit of your specific case. (Sort of like how multi-classing in Dungeons & Dragons is something you're supposed to have a long talk with the DM about.) Not every economist is both a professional academic and a professional political commentator; most are a professional at one and an amateur at the other.
(One might say that Usenet did, at one point, contain all of academia within a few of its hierarchy-roots; but there's nothing stopping random quacks from posting to those. Imagine a version of the 80s where Usenet's sci. hierarchy would only accept posts from actual science departments in universities and research institutions—but where it was still mirrored for reading everywhere else—and you'd be closer.)
Really, what I badly re-invented here, was scientific journals, but with a 140-character publication limit ;)
Or IRC channels in +v mode, but where there's only 10 of them, and each one has millions of unvoiced members and thousands of voiced members.
I don't have a solution but am mostly just curious if this kind of platform is even possible with the number of people online now. It's not like the early days of the internet where there were technical challenges that you had to overcome to make your voice heard. Now, it's so easy to say whatever anyone wants but also so easy to take ownership of the platforms for people to say what's accurate. Case in point: There are 12 users that make up the moderators for something like the top 200 sub-reddits on Reddit where 80%+ of the userbase is.
The point of "discourse domains" as a concept isn't to create echo-chambers of (correct or not) majority views. It doesn't aim to solve every problem that moderation, ignore-lists, etc. solve. The point really isn't to limit what people say at all; it's to limit how they say it, per domain (or rather, to make it clear what the proper "how" is, when you're in this space vs. that space.) It's to reconstruct a sense of there being a conversation happening within a context of particular sub-cultural discourse norms at all, vs. the current situation we're in with the Internet where every open-membership website has the de-facto discourse norms of "a protest in the public square."
As for why you'd want that—if it isn't immediately apparent: people who are enculturated to a given domain's norms, see certain things as being offensive (norm-violating), and these things differ per domain (but not really per community, given two communities both "of" the same domain.) So isolating domains from one-another, means that you eliminate "impedance mismatch" conflicts where everyone is trying their best but the difference in discourse norms makes everyone think that everyone else is acting in bad faith. It especially helps out the people for whom the norms of "a protest in the public square" are quite far from the norms of the domain they're enculturated to.
(Have you ever seen a scientist argue with a politician on Twitter? That's an impedance mismatch. Each one is trying to get different things out of the conversation, and so neither side can actually say anything without it seeming to be an act of rhetorical sabotage when seen through the lens of the other side. And further, the politician is closer to the norms of "a protest in the public square", while the scientist is further, so usually the scientist is seen by Twitter users as committing the greater faux-pas in the process.)
> This seems great in theory but how do you "verify" these types of domains without also instituting some kind of gatekeeping?
I hope it's clear, here, that the point isn't credentialism / closed-membership communities. You aren't trying to detect whether someone e.g. "is a scientist"; you're only trying to detect whether someone is best modelled as being offended by the set of things scientists are offended by.
"Verification" (in the sense of Twitter's blue checkmark) is a KYC mechanism, not a credential per se. It helps people to find a specific person on Twitter by their name. It doesn't assert that they're a member in good standing of their field; someone could be "verified" for being infamous just as much as for being famous. It just ensures that the people who "squat" on names are the people you'd expect.
The suggestions in my previous comment for verification strategies (and for linking verification to "voiced-ness" within the domain), just take this a step further: they're suggestions of processes that would be applicable to anyone who really is enculturated to a given domain's norms. They make it hard for people "uninitiated" in the domain to come in, but not by requiring a credential or a skill test; rather, they just test for something trivial that people only learn to do after first enculturating to the domain's norms.
There's no academic who hasn't written a paper—you have to write one to complete a master's and start doing original research (or, if you're an auto-didact who's not part of an academic institution, papers are even more important, as they're the way you collaborate to advance the field at all. Like PRs are in software.)
And there's no professional politician or talking head who doesn't have at least a few people willing to vouch for them. (It'd be like saying you're a professional e-sport athlete with no ladder rankings anywhere.)
None of these things require effort; they're just things you do as a matter of course as part of the process of becoming enculturated to these domains of discourse. They're like those locks in fantasy books that are opened by pushing mana into them: they don't stop bad or evil wizards, or even incompetent wizards; they just stop people who don't understand wizardry and so will probably just muck everything up if allowed inside. (Imagine a lock on a chemistry lab where you had to draw a benzene ring to get in. Same idea. They don't want you blowing yourself up in there.)
-----
Note, though, that once you've got a working domain with a cohesive set of norms, it becomes much, much easier to moderate it, because everyone in a domain will agree what the social norms they're abiding by are!
That's the real point of all this: if you untangle from your social platform all the different sub-cultural domains with their different discourse norms, then each domain can be moderated independently, by people who are also enculturated to that domain's norms and so understand them implicitly. They won't be blindly applying a checklist of rules; they'll be applying the spirit of a law that both they and the people they're applying it to understand.
(There's a good comparison to the "cultural laws" of Judaism and Islam, here. Take "cultural laws" and the priests who study+interpret+enforce them; subtract the religious aspect; and you have a domain and its moderators.)
arxiv.org
So you're saying Biden's isn't going to send me back my Bitcoin doubled?
Dang it.