Twitter Blog

Worth noting that political ads are already banned on Twitter and have been for a couple years:

https://twitter.com/jack/status/1189634360472829952?ref_src=twsrc%5Etfw

In practice, this leads to weird scenarios where Exxon can run promotional ads but climate groups can't run ads in opposition.

I'm not saying this is the wrong policy necessarily, just that "politics" is a tricky thing to define. Labor unions can't run ads saying workers should demand more pay, but Amazon can run ads saying their jobs pay good wages.

Edit: Also, I tried to find the original post but the blog is down, at least for the moment for me. https://blog.twitter.com -- anyway, here's a cached version: https://webcache.googleusercontent.com/search?q=cache:ADXc6Wf2sSAJ:https://blog.twitter.com/en_us/topics/company/2022/accelerating-our-climate-commitments-on-earth-day+&cd=1&hl=en&ct=clnk&gl=us&client=safari

I said “parts of the recommender system code.”

This is the kind of highly emotional reaction that’s not helpful.

Yes, I am quite familiar with building ML models, both training and building my own for which I’ve been paid large sums of money, and I’m here to tell you that you don’t know what you’re taking about.

There’s so much more information about an ML system than just the trained model that is important for understanding the effects of the system on a society, and its legal, ethical, and social ramifications.

Just seeing the type of RS being used, the ranking approach, and the information on SimClusters is enough for RAI folks to start to understand the ecosystem effects and how that can show up downstream in social effects.

https://blog.twitter.com/engineering/en_us/topics/open-source/2023/twitter-recommendation-algorithm

> Every profession develops their own lingo over time.

In this case, Twitter packages up two objectively observable things, people you follow and people who are popular, along with Twitter's own opinion about who the bad-faith actors are. It's that last subjective part that is at issue here. Yet all three are lumped together under the term "rank" [1], and we are told it is the bad actors who are manipulating, not Twitter.

I'm not calling for the demise of Twitter's former leadership, just calling out corporate speak where I see it. YouTube is at least transparent in this respect. They openly say that they "reduce" content [2].

[1] https://blog.twitter.com/en_us/topics/company/2018/Setting-the-record-straight-on-shadow-banning

[2] https://blog.youtube/inside-youtube/the-four-rs-of-responsibility-remove/

I don't see any calls even in the article to censor, I see a call for compulsory editorialisation.

I think structurally, if I was going to devise some means of "Fact-checking", it would be conceptually similar to birdwatch [1] because I just don't think we need this cabal of privileged elite editors to counter the stories that the masses push to the top because they're fools prone to sensationalism. One of my favorite algorithms is Reddit's "sort by controversial" because that's how you can read what a real critic thinks about something. I just think promoting traditional journalism as the cure for social media isn't the answer, a social media problem needs a social media solution.

[1] https://blog.twitter.com/en_us/topics/product/2021/introducing-birdwatch-a-community-based-approach-to-misinformation

Twitter has ads serving infra, recommendation systems (timeline, notifications, events, users), user generated events, prediction systems (ads), user graphs. The complexity is from processing and persisting exabytes of data in company owned datacenters. eg. Twitter stores images, videos, user events, user data, tweets/replies. WhatsApp has little persistence outside of metadata maybe? But your messages are not stored in a FB datacenter and if they are I'd be concerned. You can read about their infra in their blog. Comparing p2p messaging versus a distributed social media site with mountains of data and years of iteration in ML systems does not make sense.

https://blog.twitter.com/engineering/en_us/topics/infrastructure/2017/the-infrastructure-behind-twitter-scale

Why though?

It's worth noting that Twitter built their own system (EventBus) that Apache Pulsar largely mimics in design (and the people who started Pulsar at Yahoo had worked on EventBus prior), with brokers decoupled from storage, and then eventually just decided to get rid of it and use Kafka.

https://blog.twitter.com/engineering/en_us/topics/insights/2018/twitters-kafka-adoption-story.html

> One catch to this is that for extremely bandwidth-heavy workloads (very high fanout-reads), EventBus theoretically might be more efficient since we can scale out the serving layer independently. However, we’ve found in practice that our fanout is not extreme enough to merit separating the serving layer, especially given the bandwidth available on modern hardware.