ok i'm starting a fresh account in the hopes that it'll stop Discover from recommending american politics to me. if the experiment works out, i'll delete this one
i’m having so much fun with this btw. if you want to pick up math fundamentals in a rigorous way (and learn a proof checker at the same time), Tao’s Analysis book with his new Lean companion repo is the perfect start imo. you might want to practice Lean a bit first ofc but still, exciting combo
RE:
started a repo to track my solutions to Tao’s Analysis challenges in Lean. i’ve previously solved exercises up to chapter 5 or so but it wasn’t as enjoyable without formal verification. aiming to compete Section 2.2 tomorrow, here’s work in progress so far:
[Section 2.2 solutions by gaear...](
appreciate seeing the "good" concern behind my harsh phrasing. really all i want at this point from Discover is:
- actually make "Show less" work, at least for "clusterv2" segment. just check the map if i "Show less"d a cluster >2 times and always exclude it
- improve cluster mapping? mine are bad
RE:
i don’t think making Discover itself open source is a great idea, but maybe there are smaller pieces that make sense to share? e.g. if the clustering algorithm is actually good, wouldn’t it be good for the network if anyone can play with a snapshot of clusters (to make feeds, or for other tooling)?
RE:
i lowkey hate custom feeds on product level (defaults matter the most, switching between feeds is clunky etc). but, being frustrated with Discover not respecting my direct output ("show less"), i kinda see the point of being able to "walk away" to a different algorithm. if only we had alternatives!
RE:
if making the default algo (Discover) good is unachievable for whatever reason (i really don’t know why, but let’s take this premise) — i wonder if anything could be done to empower third-party feeds to serve something better? can atproto actually deliver on the promise of real algorithmic choice?
View quoted note →
