In social media land, new entrant Bluesky has been having a bit of a moment lately. When Musk declared ‘cis’ a slur, the site grew by 10%; when Musk decided to shut down Twitter over the weekend, it grew another 20%. Their corporate comms are extremely bad, so there have been a lot of weird …
Bluesky, ExplainedPost + Comments (171)
In addition to things like custom algorithms, this also means making it easy to move your account somewhere else if you don’t like the way your instance is being run. The integrity of your social graph–posts, followers, blocks, the people you follow–is ensured cryptographically, so moving it is as simple as handing somebody else the key. (This has led people to believe that there are blockchains involved, but there are not. They just share an underlying data structure.)
Compare to Mastodon, where switching your account over requires the consent of your current host, who is the source of truth for your social graph; on Mastodon you also cannot migrate your posts. More on these differences below.
Who owns Bluesky?
Bluesky, PBLLC is a Public Benefit Limited Liability Corporation that was founded in 2021. It has nine employees. The CEO is named Jay Graber; she is relatively young, and a veteran of the Distributed Web world, aka some of the non-scammy parts of “Web3”, if you remember that. The lead developer is named Paul Frazee. He is my age, also a DiWeb veteran, and a longtime advocate for free and open source software. The other employees I don’t know as much about, but I believe they’re split between PR, back-office, and protocol engineering. There is also a 24/7 moderation team.
The project that would become Bluesky began as an independent team at Twitter, where Jack Dorsey wanted to create or tailor a free and open source protocol that Twitter could use in the future as part of a push towards decentralization. As such, he sits on the board, as one of three members, but that is the extent of his involvement. The exact share ownership breakdown is unknown, but his is not a majority. (It is also not the only decentralized social network he’s been involved with; he’s more prominently backing the cryptocurrency-community-heavy Nostr, because he is kind of a moron.)
What is ATProto?
A way of communicating between instances. When anybody does anything, it is stored and broadcast as an event, which paired instances can listen to. Using Merkle trees for this theoretically makes it easy to ensure integrity and process messages as patch updates. The protocol spec also calls for external services that are dedicated to indexing and search (which can also be selected by the user, like custom algorithms). This is the “big world” mentality I mentioned, compared to Mastodon’s “small world” mentality, where there is no search engine for the Fediverse, significant friction to finding users outside your own instance, and significant pain points upon account migration. Mastodon is good at being Mastodon, but it is bad at being a federated Twitter, which is why everybody bounced off it in October/November when they attempted to migrate en masse.
One major downside of ATProto as currently designed is that almost everything is public. Everything you post, all the images, even which accounts you’ve blocked. (They’re accepting comments from devs with ideas for fixing the block part.) This information can be private in Mastodon, though administrators of instances that federate with your account do have access to it all.
How does moderation work?
Right now, bsky.social is a single instance, with pretty standard social media moderation policies. Don’t threaten to beat Matt Yglesias to death with a hammer, or torture the children of members of the Supreme Court; don’t use slurs; don’t spam; you get the idea. The response time is pretty good for a small company with 200,000 users–problematic accounts are usually taken down within forty-eight hours, though posting lurid assassination fantasies will get you nuked within the hour.
Users can create mute lists, which other users can subscribe to, which are also useful.
Adult content and violent imagery are recognized by AI and labeled as such. You can choose to show, warn, or hide this content, along with things like violence, impersonation, and ‘hate groups’.
Wait–hate groups aren’t banned?
Of course they are. Which brings us to community labeling. Federated ATProto will support third-party labeling services that you can subscribe to. There will always be discontent about what constitutes e.g. ‘sexually suggestive’ or ‘violent’ speech; if you don’t like how posts are being labeled, you can switch to (or supplement with) a different labeler.
As the largest instance, bsky.social will have significant coercive power here, but it will not be the centralized voice of god.
I heard Bluesky is anti-black and anti-sex worker.
Okay, so the sex worker thing is for sure a tempest in a teapot. Apple made the app hide all NSFW content by default. You can only change this setting on the website. Apple made Reddit do the same thing, this is just an Apple issue. As for whether a thirst trap with erect nipples poking through constitutes ‘sexually suggestive’–I weighed in on this issue (it does!) and got blocked by about fifty people who are trying to advertise their OnlyFans. Nothing of value was lost.
Anti-blackness… I’m hesitant to weigh in on this, but I think it’s being reported very badly. This TechCrunch article can give you some background, but not very well, in my opinion. There have been a couple of blow-ups that more or less ended up pitting the trans community against the black community there. Like most forum drama, it all centered around a small number of users and posts, then blew up into an out-of-control hurricane of recriminations.
Somebody even made a list of users whose block lists are more than 10% black, to call them out. Considering that a lot of this drama revolves around an old Twitter beef between a few trans people, that list ended up mostly being trans people trying to hide from who they see as their old abusers. Ultimately the creator took the list down. My favorite part of that saga is that the creator originally set it at 9%, but this ended up including an account they liked, so rather than reconsider the whole idea they just upped it to 10%.
The fact remains that the app is significantly better-moderated than Twitter ever was.
I heard it’s based on a blockchain.
No! See “Why is Bluesky?” above.
I heard Jack Dorsey owns it.
No! See “Who owns Bluesky?” above.
I heard artists who post their work there are granting Bluesky ownership or something.
Nope. They originally used an off-the-shelf Terms Of Service that had some muddled wording on this point, though it was no worse than any other social network. It has been updated to be clearer. (Basically, they can use a screenshot of you sharing your work in promotional materials–same as everywhere else.)
I heard that they’re unfair about invite codes.
100% true. Every user is supposed to get one invite code every two weeks, but this doesn’t happen for a lot of them. Other users can get big piles of invite codes for targeted community building–for example, a couple of prominent black Twitter users were given hundreds. Something similar happened with the digital artist community. I got a dozen or so to help build the author community. Communication around this is very bad! They obviously do not have a plan here.
There are many more rumors!
I’m happy to answer any questions you might have in the comments, to the best of my ability.
Promoted from the comments: I heard images you post on Bluesky go straight into generative AI training!
This is not true. Here’s a response somebody got from the team:
Hi Marco, We currently use Hive’s existing models to generate labels for images — so this is the tool that helps us identify NSFW images, so we can apply the proper labels to them. Hive does not use Bluesky account content to fine-tune/train their models. That said, Bluesky is a public social network and AI companies might be scraping data from the internet! We actually just revised our Terms of Service to have less legalese in response to conversations with artists a few weeks ago to clarify that we do not own their content. That updated version of the TOS was published last week and you can find it here: [I link this in the post]. We’re always striving to make the dense technical, legal, etc. info easier to parse. From the Twitter thread that was going around though, it seems like there is some confusion in part with Hive’s terms — Hive’s terms state that they can train on “Demo” content freely, which is what users may submit for no cost to their demo pages. “Customer Content” is a different category altogether and they do not train on that data without explicit permission from the customer, and Bluesky did not grant that permission to them. Thank you.