The openness of the ATP ecosystem is an amazing quality. Makes it feasible to extend what the Bluesky team built in a hundred ways. They made some very thoughtful design choices :)

Ordinarily bootstrapping a social platform from scratch is fiendishly hard. But with ATP, the content and creators are already here—we just need to do a better job of connecting them to consumers. The cold start problem is mostly solved for us

What’s the A/B?

- Infer topics from posts automatically
- Weight reach by user's influence in a topic, calculated PageRank-style
- Serve conversations as digest with most relevant on top instead of variable-rewards dopamine lottery feed
- Serve suggested topics to join as weekly digest
- Friendlier UX

Working on a better Reddit for ATproto! If anyone’s interested in this problem space hit me up

Interested to hear more—what’s an example of a lexicon you’re imagining here? I suspect ML remains relevant even if it’s exposed to the user via a lexicon.

Happy to chat here or elsewhere, will dm you my cell

I’m sure there are virtual gatherings but don’t know about them personally :)

Fully forgot this person existed

Thank you @knowtheory.net and @archive.org for the excellent #atproto meetup today in SF—appreciate you convening and inspiring to meet everyone else who’s building :)

Who's building / interested in a topic layer for ATP?
- Infer topics from the post corpus dynamically
- Use topics + collaborative filtering to offer better default feed
- Let user tune their topics
- Weight reach by user's influence in a topic, calculated PageRank-style

Hi Maddy - if you care to share, I'm curious to hear a high-level outline of what you'd recommend for Discover. Working on something related. Happy to chat by DM if you'd prefer.