fc-updates

Spam filtering is coming to Warpcast. 

Over the last 3 months, we've labelled thousands of spammy accounts by hand and @akshaan has carefully trained an ML model to do this automatically. 

We're rolling out this model starting today, and it will significantly improve the user experience.

Spam is posting irrelevant content in public spaces to farm reactions and follows. 

Some examples: 
- Posting irrelevant replies to someone's cast 
- Posting irrelevant content in a channel
- Mentioning or recasting users unnecessarily

Here's a one from a few minutes ago:

Technowatermelon. Elder Millenial. Building Farcaster. 

nf.td/varun

Warpcast won't show casts from spammers in public spaces. 

They won't show up in the channels, replies and notifications of people who don't follow them. 

Warpcast does promise that content will always be visible to followers (sufficient decentralization), even if a user is labelled spammy.

Our spam model is going to start labelling new accounts today, and over time will roll out to 100% of accounts. 

The labels aren’t visible at this time, though we plan to make them publicly available to other applications just like the power badge.

Artist from the future.

NFTs since 1998.

Bitcoin since 1985.

Your second favorite artist's second favorite artist.

https://manifold.gallery/tomato

the model mostly looks at message content and location to determine spam. 

if you make a cast that annoys one person, that's not really spam. 

if you make the same cast asking people to follow you in 5 high traffic channels, you'll get the spam label.

Is there a plan for how to handle false positives? I'm always worried about being labeled as a spam account because of one stupid comment someone didn't like.

How will false positives be handled? Is there manual review in the loop at all (eg for appeals?)

Can a spam-labelled account ever be labeled good in the future?

Degen

In the next three weeks, cast actions & warpcast nerfing in reply tips (im guessing they will) will kill the meta memecoin attention distribution

and so it begins
https://warpcast.com/df/0x0a3b4984

will do my best to not become a false positive food for the model

Farcaster

My Japanese CTO & Creative Director who create frames, run channels, post good content were shadowbanned today. My English speaking team member & I weren't.

My Japanese team members now don't want to use Japanese on FC.

I know there're growing pains w/ user growth, but something needs to change.

@dwr.eth @v
👇

This is one of the reasons I'm highly skeptical about spam filtering. There are going to be a lot of edge cases that the model won't handle properly, resulting in people getting unfairly shadowbanned. Most of those people will never trust warpcast again.

This is one of the reasons I'm highly skeptical about spam filtering. There are going to be a lot of edge cases that the model won't handle properly, resulting in people getting unfairly shadowbanned. Most of those people will never trust warpcast again.
https://warpcast.com/renstern.eth/0x264205b4

amazing! is "report cast" an input to the model, or thats just used for manual hiding? i've been using that more hoping it helps

This is amazing news. Excessive spam is a key reason why other platforms like X have lost their allure. Thanks team! 👍🏼

Have been doing this for web2 social networks since 2018 at SocialForensics.com

Would love to discuss with you and @akshaan if you’re game

Interesting updates! Wondering what happens to bots and other automated posting clients?

this is an interesting approach -- will there be a system in place to check if someone legit is nerfed and eventually take them out of the nerfhole?

Would you imagine a Warpcast app where you can select the algorithm that you use to curate your feed?