Post

Stealing @hailey.at's thunder to explain something cool she just did. Now when you click on a trending topic, you'll go to a screen that looks like this (1/5)

Is that a feed? Why, yes it is! In fact, if you go to @trending.bsky.app you can find the trends in its feeds! (2/5)

Hailey's creating a feed dynamically each time a trend is added. By "creating a feed" I mean she just creates a feed record on the @trending.bsky.app account. All of those feeds are hosted by the same server, which dynamically looks up the posts associated with the trend (3/5)

One interesting result of this is, we have all of these feeds created per trend. What do we do with them after the trend is over? We haven't decided 100% yet, but one interesting option is that we just freeze them as archives, becoming a snapshot of the events of that day. Which is fun! (5/5)

Is there anything in the works for putting a clear visual cue on posts that are part of a thread? I can't be the only person going bananas because people keep responding to posts without opening the context threads and a visual indicator will never solve the problem completely but it would help!

So basically- any time a post within a thread shows up, whether first or not, try to indicate it’s within a thread?

If resources were not an issue, the ideal behaviour would be to keep trending feeds with followers active forever, so people can just pin a topic to keep following it even after the trend is over. But maybe that doesn't scale?

If you DO archive them, then add a date of archival to the name, so that there aren't hundreds of indistinguishable feeds.

Woow, this is a really cool way of making it work, I really really like the idea 👏🏻🥳 Amazing work @hailey.at 🔥

I have 2 little doubts though 🤔 Is there any way for us to see the most recent posts? To know what people is talking about the topic right now. And, will people outside english speaking countries be able to se Trendings in the app soon? Because it appears for me on web but not in the app.

that’s extremely interesting for future recall, especially if there’s a way to go from a tweet to its associated archive-feeds

erm its "skeet"

Fascinating! How does this sort of semantic model work?

Is it the sort that analyzes a corpus and predicts likelihood of a match based on surrounding words? If so, I wonder what corpus it uses

Yep as I understand it (I could be wrong) she's using an embedding model, so it puts the words into a kind of semantic coordinate space where the surrounding words which are "close" will help distinguish what's being discussed