This is the clearest analysis I’ve seen of the fundamental problem with Community Notes in crisis situations: The overwhelming majority of impressions on tweets happen in the first 1-3 hours after posting, but it takes an average of 10 hours and 44 minutes for Community Notes to publicly show up. - ThreadSky

yoyoel.com • 504 days ago

This is the clearest analysis I’ve seen of the fundamental problem with Community Notes in crisis situations: The overwhelming majority of impressions on tweets happen in the first 1-3 hours after posting, but it takes an average of 10 hours and 44 minutes for Community Notes to publicly show up.

Reposted from Conspirador Norteño

I took a look at how Community Notes is doing at handling the deluge of misinformation related to the current Israel/Hamas war. TLDR: Community Notes has managed to label some popular false posts, but is falling short in multiple ways.

conspirator0.substack.com/p/community-...

Comments

nh4n3.bsky.social•504 days ago

How would you normally prevent misinformation in a time like this?

yoyoel.com•504 days ago

I don’t think prevention is the right framework; there will always be both a supply of misinfo and a demand for it. But you can take steps to add friction to its circulation and distribution — disabling reposts, adding prompts to think before reposting, and limiting recommendations can all help.

centervoices.com•504 days ago

I’m always an advocate for education and social media literacy but it has limits. Some ppl believe what the want to believe and certain unscrupulous news outlets will gladly feed it to them. Slowing down proliferation seems useful- just need to slow it down long enough for the news cycle to move on.

nh4n3.bsky.social•504 days ago

Were you able to identify it quickly and at scale?

nh4n3.bsky.social•504 days ago

Twitter is definitely worse with misinfo under the new regime but it wasn't exactly good before. I don't doubt that y'all were doing your best, this just seems like an extremely difficult problem without a good solution

slickrockweb.bsky.social•503 days ago

Yoel’s team was able to at times, deal with repeat disinformation offenders. Either through temporary or permanent suspensions. De-platforming did work. Now there are no consequences 4 just blatantly making stuff up over & over & over again, no matter how dangerous or consequential the disinfo is.

yoyoel.com•504 days ago

I assume you’re asking about my former team at Twitter. The honest answer is yes, in limited situations. Generalizing scaled misinformation detection — or hell, even detection that a post contains a claim of fact that might be misinformation — is hard. We were investing heavily in figuring it out.

bettypge.bsky.social•504 days ago

One thing I have noticed … is there’s no way to report misinformation. Theres 6 ways to report violent extreme behavior but not misinformation…

lydiabeeyoobee.bsky.social•504 days ago

There was a way to report misinformation only for Covid and elections. That was it.

joi-itos-jibo.bsky.social•504 days ago

Unrelated but your pfp rules

nh4n3.bsky.social•504 days ago

Thanks! (Need to cop to it being AI generated)

dkhunter.bsky.social•504 days ago

The prompts actually helped, huh? They felt silly, but I suppose many useful things can feel that way.

monewsdaily.bsky.social•504 days ago

Since disinfo research knows that amplification of such is the goal on TWTR, and it uses/used both many, many fake accounts (automated or human) and "supernodes" of real but complicit humans with high follower accounts...I can't believe it is that hard.

nh4n3.bsky.social•504 days ago

I assume it's easier to disable disinformation campaigns from centralized actors than organic misinformation

yoyoel.com•504 days ago

👆👆👆 (but also neither are easy, and both require sustained investment and expertise)

monewsdaily.bsky.social•504 days ago

Or...a change to the fundamentals of a platform?

"advertising" based?
"verified" users?

I mean...partly I believe it is why Elmo changed a lot of the X platform...to enable amplification and nodes.

beaverstanacct.bsky.social•504 days ago

I've always wondered how much control platforms have for vitality coefficients on posted content, and if there's a general way to pump the brakes on content to prevent it from going viral faster than people can handle, and if that preferentially slows down misinformation?

beaverstanacct.bsky.social•504 days ago

*virality coefficients, ducking autocorrect

nh4n3.bsky.social•504 days ago

It seems to me that if you've identified a post or user it's almost trivial to do that on an algorithmic platform

beaverstanacct.bsky.social•504 days ago

To an extent, but the identification at scale is the challenge, isn't it?

hoffsbeefs.bsky.social•504 days ago

Not allowing rando bad actors to promote just by paying for engagement

hammancheez.bsky.social•504 days ago

Something something lie runs round the world something before truth something boots on

jerrylevine.com•504 days ago

To be fair, it was far more effective and fast moving when it was Birdwatch and the "Birdwatch volunteers" had active communication with Twitter and NG who was managing the Birdwatch community. @conspirator0.bsky.social

All that is gone now (and I'm sure nearly all the former BWers are gone too).

ruralanarchy.bsky.social•494 days ago

What's that quote about a lie can run around the world before the truth can tie its shoes?

backrolls.bsky.social•504 days ago

You mean content moderation takes math and logic to implement?!

Who could possibly have warned E-loin…hmmm

melissaf.bsky.social•504 days ago

appreciate your insight, yoel
i'm sorry that that asshole put you and your family at risk

suestrong.bsky.social•501 days ago

The one advantage of community notes over having a Trust and Safety team is it doesn’t cost Elon 💵💵💵. Doubt he cares about misinformation as he spreads it. I spent a little time there blocking people who spread it. Waste of time. He’s destroying Twitter (X).

vanity.bsky.social•498 days ago

what if somebody lied on the computer 😳

andycraig.bsky.social•504 days ago

They’re also often of pretty low quality themselves, and often used more for an opinion clapback rather than real factual correction. The latter is what made them popular, too.

tchriste.bsky.social•504 days ago

When I was a $2.99/mo Twitter subscriber & early member of CommNotes, from the start I knew it wouldn’t work as a fact checker as too many were giving opinions or links irrelevant sites. Now the function is a joke due to blue check misuse & abuse.

colin-fraser.net•504 days ago

One of the reasons it takes so long for notes to appear is the consensus algorithm both requires a lot of ratings to work and sets a high bar for what is displayed. The result has been that visible notes are surprisingly reliable; most of the trash never makes it to “currently rated helpful”…

colin-fraser.net•504 days ago

I’m a bit worried that X will respond to complaints about the time it takes to show a note by altering those parameters in a way that lowers the bar for visibility, which will cause more of the trash ones to slip through

kuntieplopkins.bsky.social•503 days ago

kuntieplopkins.bsky.social•504 days ago

The amount of utter crap being reposted - none of it verified

Comments

Posting Rules

Reply