Realtime and search API for post embeddings! Just copy to your coding agent and it should be able to make sth out of it. Happy for any kind of feedback, especially if sth does not work. Weekend project and so on. Rate limits and physics apply. divepool.social/api/v1/openapi
Did "hate" go up on X/Twitter after they changed their content moderation rules post-Musk? What is "hate" anyway, and what data do we actually have? I've got all the charts and analysis I know you crave!
New by me for @techpolicypress.bsky.social
www.techpolicy.press/what-are-the...
the unfortunate thing about the ai conversation is one group will say "no AI, ever! all AI is bad, ALWAYS"
and then the other group says "if you say ANYTHING bad about AI, you're just a dumb LUDDITE"
and i just want us to use AI in a pro-social way that doesn't harm communities and the planet
This can still be compatible with W3C annotations, as Margi's schema is.
I think that really the way to do this is to *make the label record the annotation*. I've proposed something here:
github.com/johnwarden/o...
So I should perhaps propose an optional `TextFacetSelector` in the label record.
But even if the lexicon supported this, the margin browser plugin wouldn't know what to do with it! Since it deals with rendered web pages, not the underlying atproto records.
So for community notes, we need more of an atproto-native concept of annotations.
The lexicon could perhaps add a new type of text selector, for example:
{
"type": "TextFacetSelector",
"byteStart": 42,
"byteEnd": 67,
"exact": "the selected text"
}
Margin's text selector fields are designed for selecting a piece of a rendered web page. Selecting a piece of text *in an atproto record* is a different thing.
One mismatch here is that Margin is oriented around annotating web pages, not atproto records. You can't use Margin it to annotate a post record or a section of a long-from atproto content.
Their lexicon could support this, but it's a little tricky. A target URI could be an at:// URI, but...
But I think we could hack together a not-quite-perfect solution pretty easily to experiment.
Then we’d need to think about how labelers could annotate a section of text, not just a uri. We might look at @leaflet.pub‘s text highlight share embeds, and W3C annotations’ text quote selectors for ideas.
Hi Matthew,
Yes, I wish we had more time to talk. I think we'd have to experiment with the UI for long form. Maybe you can highlight a piece of text that you think makes a false or misleading claim, propose a note, and then other users will see some sort of icon they can click or hover over.
The people that got impatient and used the stairs missed out.
I think it’s good that it was broken. It forced more interaction.
Yes! I think it was.
Goodbye Vancouver! And the elevator at Gage Suites at UBC. I think the elevator track at the #ATmosphere conference is where I made the most interesting connections.
There is some interesting discussion in the replies to this post. I think @masnick.com com makes some good points here, but I don't agree that algorithms that just show you want you want to see are simple "conduits of speech". These innocuous-seeming algorithms have tremendous social consequences.
Could social media make us less polarized instead of more?
We tested 5 algorithms on 3 platforms with 10,000 people for 6 months during the 2024 election, and found that the answer is yes.
🧵
Come see my talk about bluenotes.social -- Community Notes for atproto -- at the Atmosphere conference in Vancouver next week!
atmosphereconf.org/event/1AzdYWM
I‘d like to know too. I’ve created community notes prototype at bluenotes.social, but this isn’t me.
The section title is "Arrests are shifting from criminals to non-criminals" (all unauthorized immigrants, criminal are not), and the text of the section supports this thesis except that one sentence which shifts unexpectedly from the criminal/non-criminal to the authorized/unauthorized axis.
You do touch on the authorized/unauthorized distinction later (restrictionists wanting to reduce immigration overall), but dropping it into this sentence breaks the logical flow.
Because that's the bait-and-switch axis your whole piece is about. The public story is “we're removing criminals,” but enforcement is increasingly hitting people with no convictions.
We are about to find if we are intelligent enough to create a computer program that is intelligent enough to create a computer program that’s more intelligent than we are.
Yes I'd love to talk to you about how your are implementing community notes in your app.
Great post as usual Jonathan.
By the way I think found a small error:
"Part of what is going on here is that a vocal Red minority want to restrict immigration generally, unauthorized or not"
I think you mean to write something like:
"unauthorized immigration generally, criminal or not".
Either ICE has started cruelly arresting beloved community members, or ICE is merely enforcing existing laws to remove murderers and rapists.
Both of these are only half-right. Whatever side you’re on, you probably have something to learn.
www.betterconflictbulletin.org/p/annoying-f...
Thanks for the suggestions! Here's the link to the post on the forum: discourse.atprotocol.community/t/lexicon-em...
Thanks @erland.sh. I posted an updated proposal to @bmann.ca Lexicon Embeds discussion.
discourse.atprotocol.community/t/lexicon-em...
I've recently been thinking along similar lines:
bsky.app/profile/jona...
@bnewbold.net , @pfrazee.com, @alex.bsky.team check out this proposal for oEmbed support in other ATProto lexicons! It builds on lexicon publishers for web embeddings. Thoughts? Should I post this somewhere? #atproto