Jasper
@raedisch
Developer from Hamburg, Germany Building
Perfectly aware that embedders will mostly embed by language rather than locality😅, I still wanted to know how much map like the 2D projections would look like. (E.g. this is Harrier 270m embedding 15k largest cities with country en/en.) More: divepool.social/analyses/cit...
There he is, @sebastian.eurosky.social, pitching Eurosky with (not against) Aya Jaff from The Valley and Felix Hlatky, director of Mastodon😱.
BTW haven’t talked once about soccer while in Vancouver. Great conference, thank y’all so much!
Got some children books in Vancouver. Two at kidsbooks, two at White Dwarf, both amazing stores.
How did this one go under my radar? Premium tv night recommendation by @gov.glados.computer, talking about how in the future, we might not have to dominate anymore, in order to succeed.
“Nobody in the room lost their job to AI yet.” - first talk of the day by @hmans.dev about woop?! coding at Agentic Conf Hamburg
Wir waren vormittags schon Rodeln und Gespräch war auch mit Spaziergang verbunden.
Visualization of this: correlation of the ratio of non-English texts to English texts to the overall mismatch percentage of detected to declared languages. The small hike in just the mismatch rate starting 2024-06-08 might have been a language code bug in the Android app: github.com/bluesky-soci...
Somewhat related: with English speaker influx, language detection mismatches drop, with non English, they do not, probably since they contain more language switchers. ~3% error rate overall. Detection via lingua-go. Cobbled together with incomplete data, so would appreciate confirmation by others.
Not a bad way to start into 2026! (Accounts are matched on all texts, but now which post is shown is taken from the latest 50 posts per matched account, which has the side effect of showing overall more recent posts, although older ones might contain better matches.)