spacecowboy
@spacecowboy17.bsky.social
Interests in ML and social aspects of tech.
Building For You feed: https://bsky.app/profile/spacecowboy17.bsky.social/feed/for-you
Hobby project: linklonk.com
Building For You feed: https://bsky.app/profile/spacecowboy17.bsky.social/feed/for-you
Hobby project: linklonk.com
Thanks for the feedback!
For You does not keep track of quote posts - it just looks to it as a regular post.
Maybe the issue here is that the same user that you are connected to likes a bunch of quote posts? In that case the solution could be to limit the amount of influence a single user has.
For You does not keep track of quote posts - it just looks to it as a regular post.
Maybe the issue here is that the same user that you are connected to likes a bunch of quote posts? In that case the solution could be to limit the amount of influence a single user has.
December 6, 2025 at 11:29 PM
Thanks for the feedback!
For You does not keep track of quote posts - it just looks to it as a regular post.
Maybe the issue here is that the same user that you are connected to likes a bunch of quote posts? In that case the solution could be to limit the amount of influence a single user has.
For You does not keep track of quote posts - it just looks to it as a regular post.
Maybe the issue here is that the same user that you are connected to likes a bunch of quote posts? In that case the solution could be to limit the amount of influence a single user has.
For your readers
December 6, 2025 at 12:55 AM
For your readers
It gets you to consider whether this post is aligned with the goals of your future self.
The slight local change in behaviour may create a system level change.
Maybe it would create an incentive for more informative content over opinion warfare.
The slight local change in behaviour may create a system level change.
Maybe it would create an incentive for more informative content over opinion warfare.
December 5, 2025 at 3:01 AM
It gets you to consider whether this post is aligned with the goals of your future self.
The slight local change in behaviour may create a system level change.
Maybe it would create an incentive for more informative content over opinion warfare.
The slight local change in behaviour may create a system level change.
Maybe it would create an incentive for more informative content over opinion warfare.
Yes, For You changes the meaning of a like. Instead of being purely an outbound message (increase the counter for others to see, notify the author) it becomes a signal that determines what your future self will see in For You.
December 5, 2025 at 2:47 AM
Yes, For You changes the meaning of a like. Instead of being purely an outbound message (increase the counter for others to see, notify the author) it becomes a signal that determines what your future self will see in For You.
Yes, For You only uses your likes and "show less"/"show more". Maybe they meant Discover? I don't know if Discover uses clicks on posts/profiles.
December 5, 2025 at 2:34 AM
Yes, For You only uses your likes and "show less"/"show more". Maybe they meant Discover? I don't know if Discover uses clicks on posts/profiles.
No, can you describe the use-case?
The debug site shows recommendations based only on the user's public likes.
When you access your feed in-app your feed is affected by what you have seen in the past (seen posts are filtered out) and by your "show more"/"show less" interactions.
The debug site shows recommendations based only on the user's public likes.
When you access your feed in-app your feed is affected by what you have seen in the past (seen posts are filtered out) and by your "show more"/"show less" interactions.
December 4, 2025 at 12:17 PM
No, can you describe the use-case?
The debug site shows recommendations based only on the user's public likes.
When you access your feed in-app your feed is affected by what you have seen in the past (seen posts are filtered out) and by your "show more"/"show less" interactions.
The debug site shows recommendations based only on the user's public likes.
When you access your feed in-app your feed is affected by what you have seen in the past (seen posts are filtered out) and by your "show more"/"show less" interactions.
That was quick, thanks!
What is it sorted by now? It is not by likes, right?
What is it sorted by now? It is not by likes, right?
December 4, 2025 at 1:41 AM
That was quick, thanks!
What is it sorted by now? It is not by likes, right?
What is it sorted by now? It is not by likes, right?
num users - is the number of people that have liked at least one post as you and that also liked the recommended item
num paths - is the number of paths that connect you to the recommended item X through: item A you liked -> user B who liked item A -> user B liked the X
popularity - # post likes
num paths - is the number of paths that connect you to the recommended item X through: item A you liked -> user B who liked item A -> user B liked the X
popularity - # post likes
December 4, 2025 at 1:15 AM
num users - is the number of people that have liked at least one post as you and that also liked the recommended item
num paths - is the number of paths that connect you to the recommended item X through: item A you liked -> user B who liked item A -> user B liked the X
popularity - # post likes
num paths - is the number of paths that connect you to the recommended item X through: item A you liked -> user B who liked item A -> user B liked the X
popularity - # post likes
The suggested feeds used to be generated dynamically based on the likes table: github.com/bluesky-soci...
github.com
December 4, 2025 at 12:40 AM
The suggested feeds used to be generated dynamically based on the likes table: github.com/bluesky-soci...
Here is where it is being queried: github.com/bluesky-soci...
github.com
December 4, 2025 at 12:39 AM
Here is where it is being queried: github.com/bluesky-soci...
The "suggested_feeds" table has not been updated in >1 year and new feeds like For You don't appear on bsky.app/feeds
Please consider updating it with a fresh list of feeds.
Please consider updating it with a fresh list of feeds.
bsky.app
December 4, 2025 at 12:38 AM
The "suggested_feeds" table has not been updated in >1 year and new feeds like For You don't appear on bsky.app/feeds
Please consider updating it with a fresh list of feeds.
Please consider updating it with a fresh list of feeds.
Glad you liked it!
This debug site shows the For You feed for any account and it can explain why each post was recommended: linklonk.com/bluesky
This debug site shows the For You feed for any account and it can explain why each post was recommended: linklonk.com/bluesky
Bluesky "For You" feed playground
linklonk.com
December 4, 2025 at 12:27 AM
Glad you liked it!
This debug site shows the For You feed for any account and it can explain why each post was recommended: linklonk.com/bluesky
This debug site shows the For You feed for any account and it can explain why each post was recommended: linklonk.com/bluesky
The images don't seem to come from the reports. Did you make them with Gemini 3?
December 2, 2025 at 1:20 PM
The images don't seem to come from the reports. Did you make them with Gemini 3?
Try For You as an alternative to Discover
December 2, 2025 at 1:05 PM
Try For You as an alternative to Discover
Try For You to find content related to what your like
December 2, 2025 at 1:02 PM
Try For You to find content related to what your like
Just in case you want something more scrollable - try For You
December 2, 2025 at 12:59 PM
Just in case you want something more scrollable - try For You
Try For You as an alternative to Discover
December 2, 2025 at 12:55 PM
Try For You as an alternative to Discover
You can unpin the Discover feed - press # in the top right to get to your feed list, press the gear button in the top right to enter the edit mode and unpin Discover.
December 2, 2025 at 12:52 PM
You can unpin the Discover feed - press # in the top right to get to your feed list, press the gear button in the top right to enter the edit mode and unpin Discover.
Try For You - a personalized feed based on your likes. It shows a lot of FE content based on your likes.
December 2, 2025 at 2:06 AM
Try For You - a personalized feed based on your likes. It shows a lot of FE content based on your likes.
Try For You - a personalized feed based on your likes
November 30, 2025 at 11:32 PM
Try For You - a personalized feed based on your likes
I suggest liking more varied content so For You connects you to more people through your shared likes.
November 30, 2025 at 11:02 PM
I suggest liking more varied content so For You connects you to more people through your shared likes.
But you can't do that with floating point numbers. a + b - a - b is not exactly 0.0. It would be very close (either negative or positive) but not exactly 0.0.
To fix this I'm counting the number of co-likes minus the number of cancelled co-likes from "show less" and check if we have >0 left.
To fix this I'm counting the number of co-likes minus the number of cancelled co-likes from "show less" and check if we have >0 left.
November 30, 2025 at 5:13 PM
But you can't do that with floating point numbers. a + b - a - b is not exactly 0.0. It would be very close (either negative or positive) but not exactly 0.0.
To fix this I'm counting the number of co-likes minus the number of cancelled co-likes from "show less" and check if we have >0 left.
To fix this I'm counting the number of co-likes minus the number of cancelled co-likes from "show less" and check if we have >0 left.
For example, if there are 2 co-likes and 1 "show less" then we treat that "show less" as a - "subtract the weights of 2 most significant co-likes". End then I would skip the co-rater if the positive and scores completely cancel out each other: "if corater_weight == 0.0 then skip".
November 30, 2025 at 5:09 PM
For example, if there are 2 co-likes and 1 "show less" then we treat that "show less" as a - "subtract the weights of 2 most significant co-likes". End then I would skip the co-rater if the positive and scores completely cancel out each other: "if corater_weight == 0.0 then skip".
The bug was the classic "don't compare float values for equality". When I calculate the score contribution of each user we get the sum of weights of all co-liked items and then apply the "show less" penalty by cancelling some of those co-likes.
November 30, 2025 at 5:05 PM
The bug was the classic "don't compare float values for equality". When I calculate the score contribution of each user we get the sum of weights of all co-liked items and then apply the "show less" penalty by cancelling some of those co-likes.