Ryan J. Gallagher
@ryanjgallag.com
4.6K followers 970 following 440 posts
Applied scientist trying to make the internet a little better. PhD. Trust & safety, platform manipulation, networks, fingerstyle guitar. I use my hair to express myself. He/they
Posts Media Videos Starter Packs
Pinned
ryanjgallag.com
With more people migrating to Bluesky (and my general quietness since Twitter fell apart), I want to reintroduce myself!

I'm an applied scientist in online trust & safety. I specialize in detecting coordinated/harmful/deceptive networks, and bridging domain expertise with machine learning systems
Reposted by Ryan J. Gallagher
annebdh.bsky.social
Want to work on open source full time? The @roost.tools engineering team is starting to hatch! Come build OSS tools making a difference in Trust & Safety.

This is a fully remote role, though some schedule overlap with North American time zones is expected.

www.linkedin.com/jobs/view/43...
ROOST.tools hiring Staff Software Engineer in United States | LinkedIn
Posted 8:52:18 PM. About ROOSTROOST is a community effort to build scalable and resilient safety infrastructure for…See this and similar jobs on LinkedIn.
www.linkedin.com
ryanjgallag.com
what could go wrong with talking about moderation on substack
ryanjgallag.com
Of course, people don't agree on what is "safe" and "fair" and that's why moderation is hard. But you can't solve a network-level problem with individual-level tools
ryanjgallag.com
Moderation isn't simply about taking things down, it's also about building a collectively safe space. Telling someone they can just "label" (or hide/block) something they don't like won't address their concern that their community isn't safe
ryanjgallag.com
Do I start a blog simply to write why "labelers" aren't enough?
Reposted by Ryan J. Gallagher
boozybadger.lawyersandliquor.com
Dunno man, seems a little misleading to write an article about Trump describing activity in Portland and Chicago as “insurrection” with a 33 year old photo of a burning donut shop taken in the 1992 LA Riots (sans caption explaining that) as the header on all your posts about it.
The picture from
The article with a picture of an armed men standing in front of a burning donut shop. You must click through to read the caption “
Two National guardsmen stand guard outside a burning donut shop in Los Angeles on April 30, 1992.
The National Guard was called in to aid police during the second day of rioting in the city.”
Reposted by Ryan J. Gallagher
esqueer.net
This is the Libs of Tiktok problem that Yoel Roth wrote about and the problem of legal formalism in content moderation.

This isn't some new issue to moderation teams. It's a well-known problem.

www.lawfaremedia.org/article/cont...
Reposted by Ryan J. Gallagher
victorerikray.bsky.social
The media trots out the "deplorable" comment from a decade ago to castigate the left. Once a week now Trump implies that the military should destroy democrats, and it is just how things are.
atrupar.com
Trump to the Navy: "We have to take care of this little gnat that's on our shoulder called the Democrats."
Reposted by Ryan J. Gallagher
grudgie.bsky.social
Look. Invading our own country is about as straightforward a case as you can have for impeachment. The GOP abandoning the Constitution is kind of the only story there is and it gets no coverage.
sethcotlar.bsky.social
Guys. I just had a crazy idea. What if the President DIDN’T deploy troops in American cities. Just a thought. Have we tried that before?
Reposted by Ryan J. Gallagher
bnewbold.net
for atproto devs and protocol watchers, I published an overview of the network moderation architecture.

it tries to cover all the mod actions possible for each service type. this design has been around a while, but not well documented.

this doc is not very polished, but could clarify some details
AT Moderation Architecture | bryan newbold
The AT network is becoming more heterogeneous in practice, with independent PDS hosts, apps, and alternative bsky AppViews establishing themselves. This means that more complex inter-service moderatio...
whtwnd.com
Reposted by Ryan J. Gallagher
bnewbold.net
a bunch of folks are wondering about how mod comms work for other PDS instances. this is also a problem for, eg, labelers trying to reply to appeals.

the plan is this proposal, which we have been making slow progress towards (it has been a very very busy 12 months for mod eng team)
Reposted by Ryan J. Gallagher
anildash.com
The tech-centric solution to “everybody is always yelling at us” is “let’s make it so they can run their *own* platform”, with the idea that they can then say: If you don’t like it, go run your own service with rules you like better. That’s literally one of the design goals Bluesky was made for.
ryanjgallag.com
"but Ryan the whole point of ATProto is that you can just do it, why don't you???"

because I work a full time job and I don't love coding enough to do it 24/7
ryanjgallag.com
I think Bluesky is building dated approaches to moderation and resolution, compared to how modern the other aspects of the protocol feel

It feels like they pull in the latest tech innovations for the protocol, but none of the latest cutting edge ideas for moderation and trust & safety governance
Reposted by Ryan J. Gallagher
notalawyer.bsky.social
i actually prefer it when the mods actively hate the user base and vice versa, that’s how every great message board worked for years and i think we should give it a whirl
Reposted by Ryan J. Gallagher
anildash.com
There are a few challenges here, though. Building an ecosystem of alternate moderation systems requires third parties — both devs _and_ moderation experts — to trust in you all as both stewards of the infrastructure for that ecosystem *and* operators of the biggest/default instance of that system.
ryanjgallag.com
I get that it's a different skill set than the engineering that's been needed to get things running to date, but I think that's more worthy than going down the ambiguous rabbit hole of "healthy discourse"
ryanjgallag.com
I think Bluesky is building dated approaches to moderation and resolution, compared to how modern the other aspects of the protocol feel

It feels like they pull in the latest tech innovations for the protocol, but none of the latest cutting edge ideas for moderation and trust & safety governance
ryanjgallag.com
With respect to The Discourse™, I think the biggest thing is fairness, transparency, and accountability with moderation. I think the cloak and dagger form of moderation of social media past where minimal evidence and recourse is available to navigate takedowns leads to inequities and confusion
ryanjgallag.com
I'm not saying it's bad to aim for "healthy discourse," but this is literally what Twitter and Facebook tried in response to the 2016 US election, and it didn't really solve any core problems
ryanjgallag.com
Trying to engineer "healthy discourse" is literally a 2018 mindset of social media. It's dated, and it doesn't get to the crux of what people want or need from online spaces
Reposted by Ryan J. Gallagher
osakanone.bsky.social
Friendly reminder:

Social contagion is a network modelling tool, not a material phenomenology. They don't exist.
Reposted by Ryan J. Gallagher
juddlegum.bsky.social
If the Democrats don’t cave on the government shutdown, Trump will fire lots of federal workers (which is also exactly what Trump did when the government was open)
Reposted by Ryan J. Gallagher
julietshen.bsky.social
Tools like osprey have been around and used by T&S teams for years, and we're super excited to keep improving on the v0 that's currently out!

Prior art in this space (that's public) includes Pinterest's guardian tool

medium.com/pinterest-en...