>Second, we will prevent “liked by” and “followed by” recommendations from people you don’t follow from showing up in your timeline and won’t send notifications for these Tweets. These recommendations can be a helpful way for people to see relevant conversations from outside of their network, but we are removing them because we don’t believe the “Like” button provides sufficient, thoughtful consideration prior to amplifying Tweets to people who don’t follow the author of the Tweet, or the relevant topic that the Tweet is about.
I want to applaud this, because it's clearly a good move. but i don't see how they can square "our recommendation engine is harmful enough that we need to disable it to protect the security of an election" with "we're gonna turn it back on in four weeks"
>>Second, we will prevent “liked by” and “followed by” recommendations from people you don’t follow from showing up in your timeline and won’t send notifications for these Tweets.
I don't know if a Twitter PM will read this, but showing Likes in my feed — and not having an option to turn it off in the official mobile app client — is one big reason I stopped consuming tweets.
It's a shame that the only way to opt out of seeing Likes is by using a third-party client.
This, and other timeline muddling they've done, are the reasons I use third-party clients like Tweetbot almost exclusively (other than when clicking a link opens the official app).
The thing that irritates me the most is that they also seem to track the last time you logged in via the Twitter website or official client, and show me "while you were gone" tweets that it thinks I might be interested in. Typically, they're right - so right, in fact, that I've often already replied to them.
I've always said (until Twitter went full asshole, which they're now reversing course on) that I'd pay for a "Twitter Pro". Let me use a third-party client to do everything that the official client does, but with the option of turning things off - provide likes in my timeline, which my client can filter out; allow me to access polls via the API (which presumably they don't do for everyone because of botting), and so on.
Twitter, I will pay you monthly to stop making me choose between an incomplete experience (third-party clients) and an awful one (inflexible first-party clients). It's guaranteed income. Please, sign me up.
I don’t understand why social media companies don’t offer a paid tier. They don’t have to charge everyone as they’d lose the vast majority of users, but for those who want to pay, why prevent them? They’d be making more money than they do from ads.
Wow, that sounds really annoying. I’ve used Tweetbot exclusively for years (I’m mostly a consumer, though — I very rarely tweet), and had no idea this was even a thing.
Twitter is really working hard to force people away from using their app.
You are right to say "for now" - the app will periodically force you back to "Home" view (after a seemingly random period) and then you have to manually switch back - I'm guessing Home view has better "engagement" metrics for the Twitter suits.
I believe they stopped doing that. The verbiage in the dropdown indicating that they'll switch back has disappeared, and my feed has been locked into the latest view for months now.
At least the Twitter client on Android lets you switch to the "Latest Tweets" timeline (via that "Star" icon in the top-right corner - same as on the web client) which doesn't have all the "engagement features" of the "Home" timeline (like showing "xxx liked"). IFIR twitter automatically switched back to the Home timeline after a little while, but eventually it seems to learn to stay on the Latest Tweets timeline.
This is my recurring issue when I see most "changes we're making for the election" posts. Sure this seems like a good step for this election.
Fake news and the destabilization it can bring isn't just an American phenomenon though. Are they going to apply these precautions to elections in Brazil, or Myanmar?
For anyone else still confused: CVS is an American retailer that typically issues receipts with additional advertising, bad low-value coupons with paragraphs of exclusions, on the bottom of a receipt. This makes a single-item purchase result in a receipt exceed 2m in length.
"This makes us money but harms the world as a side effect, but if we harm it too much the retaliation will cost us money. As a result we're going to only use it as much as we think we can get away with."
> I want to applaud this, because it's clearly a good move.
It’s not that clear to me that this is a net good move. Exposure to out-network content, depending on what is exposed, could be good for preventing echo chambers and confirmation bias too. Twitter likely went the other direction and promoted out-network tweets that only increased engagement, and blaming the feature itself for this is naive at best.
Second issue everyone seems to be missing is the potential partisan-ness of these feature changes. Do the usage of any individual feature distribute equally among all demographics? If not, feature-shaping a near election could be construed as an underhanded way to hamper the discourse of certain demographics. Of course this is very easy to dismiss as conspiratorial thinking, but I think ideally any intellectually honest discourse engine would have gone the extra mile to demonstrate the neutrality of their election specific changes.
That is just false equivalency. Timing and intent changes everything. If a feature was not purpose built but happened to be used more by a certain demographic, that is not a partisan move at all, but removing it right before a political event intentionally would definitely be.
To be clear, I am not saying this is necessarily the case, but not being in the know is problematic as far as health of democratic processes is concerned, especially when public discourse more and more depends on private enterprises. I don't want any fingers on the scale, because even if it might be favoring my side today a) that is not honorable b) there is no guarantee shoe won't be on the other foot tomorrow.
I find it suspect that they are giving as one of their criteria for calling the election as two authoritative news sources independently calling it. If this were non partisan they would have at least said a majority of authoritative sources or some other consensus.
Then you run into the problem of "What is authoritative?" By limiting it to just two, Twitter can say, "We like these guys. They're reputable." Instead of having to line up and vet thousands of other sources, and task people with monitoring all those thousands of authoritative sources around the world.
I’ve never understood the dewy beats Truman calling system. In the UK each constituency announces results after being counted (and recounted). Those results arrive during the night, polls close at 10, first results at 11, a good idea of who will win in most elections by 5, andthe last few results might stretch into 9 or 10 am if ferries are having issues.
But aside from a national exit poll at 10pm, there are no real guesses.
It seems that different US elections predict who will win a given state far earlier in the process - before the votes are counted even the first time. Is that right?
My own global broadcaster takes its “calls” from ABC. I think that abc, nbc, cbs, Fox, cnn, New York Times and Washington post are probably good enough for “calling it”. I was working election night 2012 in our Washington office and was shocked how early the “Obama wins” straps went up -I have a photo at just 23:19 eastern from fox saying Obama was re-elected. That’s something like 4 hours before polls close in Alaska and Hawaii!
If 4 of those networks and papers have called it then it’s likely good. Sure there’s buzzfeed and whatever that may have enough professional staff on to really judge it, but it’s unlikely that 3 networks and a major paper are all going to call Dewy rather than Truman.
> It seems that different US elections predict who will win a given state far earlier in the process - before the votes are counted even the first time. Is that right?
Yes.
At least in Dewey / Truman, they counted the votes and Truman won. Under the current system, Truman would be expected to concede gracefully after somebody predicted he would lose, removing the need to count the votes entirely.
I really disliked that "feature", I usually use the chronological mode to avoid seeing them.
If I wanted to see that content I would follow it, and if I wasn't and the content was good someone I follow might retweet it. And if I follow someone who retweets annoying things I can disable retweets on that person.
The chronological mode doesn't help either. That's the reason why I keep using half-broken third-party clients on mobile and Tweetdeck on desktop.
If I wanted to "discover" or "explore" something, I'd open the dedicated section, thank you very much. Modern social media is ridiculously user-hostile.
> we are removing them because we don’t believe the “Like” button provides sufficient, thoughtful consideration prior to amplifying Tweets to people who don’t follow the author of the Tweet, or the relevant topic that the Tweet is about
It's amusing that it took the election for them to finally admit this. I had thought much the same thing of Twitter's recommendations for a long time. People discussing divisive political issues shows up on my feed a lot, despite deliberate curation of who I follow, just because the people whose personal work I appreciate also happen to participate in political discussions in addition to posting their work, so the issues they "like" get posted on my feed. All of the posts they liked were from accounts I had no interest in following. Personally this change would cut down on the signal to noise ratio for me.
I am thinking they had to balance the improvement in their metrics for engagement they would gain by putting more novel out of network tweets in view, and the virtol and divisiveness the election-related media and misinformation propagating further would inevitably cause. Yes, there is not enough "sufficient, thoughtful consideration" put into the recommendations, but if they ultimately increase Twitter's user engagement and revenue then it takes a lot of effort to backtrack on them. It also makes me continue to believe that some kinds of information are detrimental in some ways, and Twitter is now controlling the flow of that information as to prevent public backlash or some other harm.
I’ve noticed that they’ve replaced this with surfacing tweets from your lists. This is really annoying to me as I’ve moved political follows to a politics list so that I’m not inundated with politics and have to explicitly check the list to see that content. Now I get it all the time.
Their recommendation engine is not harmful in all cases. I used to use Twitter to follow some machine learning/programming profiles and remember the recommendations being quite good.
in my experience, if anybody you follow ever interacts with anything political, that quickly drowns out any other interesting content in the recommendations. If you exclusively follow people who exclusively post and interact with programming content, then maybe your recs will be okay. but all it takes is one person in your network to interact with one other topic to poison your recommendations for good.
This is the amazing duality of social networking: as soon as it becomes topic specific, the quality of discussion improves in direct corellation to how technical the topic gets. The moment the conversation is just general babble, anonymous / pseudonymous people will find things to argue about.
I sadly don’t find this to be true. The caustic subjects in all disciplines get talked about more and recommended more. I made the mistake of watching some flight videos and now get recommended crash montages all the time. I watched the C++ con talk on “OO considered harmful” and now have an inordinate amount of bullshit in my feed. It feels kind of crazy that my YouTube recommendations took such a downturn suddenly.
Yup. I'd love to live inside my own bubble filter.
Does anyone do personalization well?
I worked on team that was trying to transition from recommendations to personalization (fashion retailer with ecommerce, think Macys Nordstrom etc).
Our imagined gold standard was the expert sales person, the personalized shopper. So somehow figure out how to make a digital sales person. Like StitchFix or Trunk Club. Curation at scale.
But I couldn't even figure out how to solve my own ultimate shopping challenge: recommend a quality white t-shirt that fits.
Though I doubt it, maybe StitchFix has cracked this nut. If they (or someone) has, we need to distill their magic and apply it more generally.
(I haven't tried Spotify or Apple Music. I've heard their personalization efforts are pretty good, so I'm semi-curious.)
--
Oh, and a post script, while I'm chewing on this topic.
I have a theory why quality personalization isn't likely to become the norm. I can't yet imagine how it'd displace today's biz models. Targeted advertising has sucked up all the oxygen.
It's hard to articulate the difference between personalization and targeted advertising (recommenders). One of those torturous endless discussions our teams couldn't escape.
"Personalization" is needs fulfillment whereas targeting is attention stealing. Personalization is much harder to monetize, because it relies on conversions. Whereas with targeting, the money changes hands before hand and is easier to pretend it's working.
I will clarify: I don't mean that _feeds_ are doing well with this. I just mean, when you dig down to a place where Real Discussion is happening, like a web1.0 message board or group or whatever, there's still tons of good, good-faith discussion happening. It's just not the stuff that percolates up to the top. Relying on your feed recommendations for your content is like living off of nothing but McDonalds, it's not good for you.
> when you dig down to a place where Real Discussion is happening, like a web1.0 message board or group or whatever, there's still tons of good, good-faith discussion happening
I mean topic specific in the sense of finding places where deep discussions are being had - like, say, a mailing list or forum for a very particular thing, scientific or programming or cars or whatever you like. Not what you see on Facebook. Social networking is not just FAANG, look in the long tail for the good stuff.
I don't think its that illogical to argue there's a time sensitive component. False news is the price you pay for breaking news sometimes. Its fine in the long run, but the argument is during an election the risk is too great.
Do you also find it hard to square the statement "large gatherings are dangerous enough for spreading diseases that we need to ban them for public safety" with "we're gonna have them again when the pandemic is over"?
This is the truth. However, bringing attention to it and shaming them over it is still worthwhile since, while their actions are in line with the desire of their customers, it's contrary to the desire of the majority of their user base.
Recommendation engines are neither inherently good or harmful -- think of them as an amplifier.
In normal times, they work pretty well -- it gives you more of the stuff you like, which usually isn't harmful. Just entertaining/fluff/interesting/whatever.
But closer to an election, they can be gamed and weaponized for harm -- Russia can activate botnets, fake accounts, etc. and lies spread like wildfire.
So it's not about whether the recommendation engine is good or bad -- it works. The point is that close to an election, bad actors weaponize it 1,000x more or more often than they do the rest of the time, so they're turning it off. When it's not being weaponized so much, they can keep it on.
It's like paying for a DDoS protection layer at times when you know your site is likely to be attacked, but not using it the rest of the time.
It seems hilariously naive to think that the only time people will be weaponising it is during an election and that there aren’t elections happening regularly all over the world so that if your point holds true such things should basically always be turned off.
They didn't say that though. They're saying the consequences are worse before an election. At other times the pros outweigh the cons. Not sure I agree but its a pretty reasonable argument, I think.
> But closer to an election, they can be gamed and weaponized for harm
But there's elections all the time. There's many different elections in the US alone, any many many more throughout the world.
And elections are just one thing, what about influence over opinions for bills/law/reforms? Wars? Public opinion about basically any subject, from economics to gender/race/class tensions?
If you accept that social media is weaponised for this election, it seems naive to not accept it is being weaponised absolutely all the time
I want to applaud this, because it's clearly a good move. but i don't see how they can square "our recommendation engine is harmful enough that we need to disable it to protect the security of an election" with "we're gonna turn it back on in four weeks"