spacecowboy

@spacecowboy17.bsky.social

5144 Followers 59 Following

Interests in ML and social aspects of tech. Building For You feed: https://foryou.club https://bsky.app/profile/spacecowboy17.bsky.social/feed/for-you Hobby project: linklonk.com

Feed search results
For You in #7
Screenshot
The "cold-start" screenshot of For You: a list of interest-posts that new users are encouraged to like to get started.
For You with the pinned welcome post marked as pinned.
The list of Bluesky interests
A screenshot of the Feeds list with For You at 7th position
A new 96GB DDR5 RAM kit
Twitter reply
A screenshot showing the list of likes on foryou.club/profile/likes
A screenshot with A/B test metrics
For You feed with 49999 likes
Retention graphs per cohort showing ~50% at 4 weeks.
A graph of per-quarter user numbers of For You.
For You feed description screenshot
A screenshot of the stats page: Your Posts in For You Your posts were shown in For You 4,684 times and received 51 likes. Followed by graphs of views and likes by day.
A setting from foryou.club/playground: Popularity penalty: 0.3 When >0, penalizes generally popular posts: score = score / power(popularity, popularity_penalty) Bigger value => less popular posts get to the top. Try 0.1 - 0.5.
A screenshot of X post "What’s the mood on BlueSky?" which includes my reply: "The most popular post today in the For You feed has nothing to do with politics:" followed by a screenshot of a popular Bluesky joke post.
18:58–18:59 — Tailscale starts showing derp-26 does not know about peer [gXd5W] warnings (intermittent relay issues) 19:00:24 — Tailscale drops connections from VPS (100.122.235.19) to gaming PC (100.125.166.46): connection was refused — dispatch on VPS can no longer reach gaming:8090, returns 500s to nginx 19:04:14 — Tailscale closes stale derp-26 connection (age 10m59s) 19:04:15 — Tailscale reconnects to derp-26 ~19:08–19:09 — traffic resumes (feed logs reappear in parquet)
My reply to an X post: The default Discover feed might give a very one-sided impression of the platform. I built custom For You feed that is based on your likes. Here is how to find good content: - search for something you are interested it - like it - open "For You" to see more content like that
A settings control labeled "Seed recency min weight" with a value of 0.1. The description reads: "Reduces the influence of recent likes. The most recent like gets this weight; the oldest like gets 1.0; in between is linear. 0 = disabled, 0.1 = mild dampening (e.g. most recent like counts 10x less than the oldest)." Below the text is a line chart showing weight on the y-axis (ranging from 0.10 to 1.0) and time on the x-axis (from "recent" on the left to "older" on the right), with a straight diagonal line rising from 0.10 at the recent end to 1.0 at the older end.
Raw results of an A/B test for the For You feed. The main metrics are summarized in the thread text.
Dear customer, Our network monitoring system detected unusual traffic patterns that may indicate your services with IP 149.56.134.145 are under attack. To safeguard your business and eliminate any malicious activity, we will now be routing your traffic through our Network Scrubbing Centres. Please note these kinds of attacks are typically aimed at interrupting services and should not pose a threat to the security of your data. Usually, no action is required from your side.
Checkbox for "Random sample source ratings Instead of using the 500 most recent likes as seed, randomly sample 500 from your 5000 most recent likes. Makes recommendations more diverse."
uptime 99.6%
Screenshot showing "The Most Popular Feeds on Bluesky" ranked by total likes. Three feeds are listed: 1) "For You" by spacecowboy (41,101 likes) - a personalized algorithmic feed based on co-likes, 2) "Popular With Friends" by Bluesky (40,669 likes) - shows popular content from followed accounts, and 3) "Discover" by Bluesky (38,496 likes) - displays trending content from personal network.
Users with 1+ visits: 209,195 users with an 18.97% like rate. Users with 2+ visits: 155,712 users with a 24.09% like rate. Users with 10+ visits: 114,178 users with a 28.32% like rate. Users with 100+ visits: 47,830 users with a 35.01% like rate. Users with 1000+ visits: 5,509 users with a 43.06% like rate. Users with 10000+ visits: 257 users with a 36.19% like rate.