Linda Yaccarino is leaving X

(www.nytimes.com)

Show context

jeffbee ◴[09 Jul 25 15:09 UTC] No.44510975[source]▶

Have any of the people who noisily joined X to make a big impact fast actually had a big impact over any time frame? Remember when G. Hotz said he was going to fix Twitter search in 6 weeks, and then it turned out that G. Hotz is just another midwit like anyone else and Twitter search is still as bad as ever? Yaccarino said they were going to transform Twitter into the "everything app" with payments, marketplaces, and even banking. None of which it turns out was within the abilities of Linda Yaccarino.

replies(6): >>44511404 #>>44511511 #>>44511521 #>>44511529 #>>44511569 #>>44516934 #

1. UltraSane ◴[09 Jul 25 15:38 UTC] No.44511404[source]▶

>>44510975 #

Search is a pretty solved problem if you are willing to invest the resources to create a inverted index of all the text you want to search. An inverted index of all tweets would be pretty expensive. Creating text embeddings for semantic search would be the next stage and even more expensive.

replies(3): >>44511481 #>>44511813 #>>44512194 #

2. lokar ◴[09 Jul 25 15:44 UTC] No.44511481[source]▶

>>44511404 (TP) #

Basic term based retrieval has been solved for 30+ years

The problem is ranking and relevance

replies(1): >>44511556 #

3. lokar ◴[09 Jul 25 15:51 UTC] No.44511556[source]▶

>>44511481 #

Thinking more, I imagine each post has limited value for ranking. You need the context of the thread, re-posts, even other threads nearby in time (with the same people).

4. simonw ◴[09 Jul 25 16:15 UTC] No.44511813[source]▶

>>44511404 (TP) #

They've had an inverted index of all tweets since 2008 (when they acquired Summize).

They added a vector index a year and a half ago for a "see related tweets" feature - https://x.com/elonmusk/status/1720314092269822242 - though as far as I can tell that feature doesn't exist any more, presumably replaced by the ask Grok button.

5. phillipcarter ◴[09 Jul 25 16:46 UTC] No.44512194[source]▶

>>44511404 (TP) #

It is very much not a solved problem. Because the implication behind search is not "well the result you need is technically in the result set", it's "the result you need as at the top", and that remains an extremely difficult problem for anything but a trivial scale.

replies(1): >>44512498 #

6. UltraSane ◴[09 Jul 25 17:10 UTC] No.44512498[source]▶

>>44512194 #

Good support for regex and boolean operators helps a lot with that. But that requires user skill.

replies(1): >>44514814 #

7. HeyLaughingBoy ◴[09 Jul 25 21:15 UTC] No.44514814{3}[source]▶

>>44512498 #

Then it's not a solved problem in any meaningful way.

↑