Tildes and Hacker News are my go to sites for general conversation, and information and I find being largely text based is what keeps the quality of the sites from devolving. Are there other similar sites?

I recently made a post here on Tildes in ~food about a pizza I made, and linked it to an Imgur album showing part of the process. This seemed interesting to me, and didn't think of it as an image post per-se. While most of the responses were positive, and we talked about pizza-making, it clearly did upset some people who viewed it as an image-only post.

Thinking through things, image only posts can be a bad thing - but there are plenty of good reasons to make image posts as well. Would images always be ok if they are original content? Certainly doesn't completely eliminate the possibility for people to shitpost, but does reduce it considerably.

Would it be acceptable to allow image posts, but require a certain number of words/characters attached to each post? This seems like it would be pretty easy to implement, and forces the user to actually make some effort, as opposed to just "karma farming".

AND ANOTHER THING!

Subscriptions. I would really like to see more specialized groups/communities here, and the current implementation I see is encouraging. ~games now has sub-groups, ~games.tabletop and ~games.gamedesign. This same system could be extended to sports. ~sports.motorsport.formula1, ~sports.esports.leagueoflegends, etc.

I notice this primarily with the YouTube videos. I've started to notice that the videos I see posted in here I have already had recommended to me by YouTube. And I realize it must be because when I watch a video here, the YouTube algorithm decides I'm interested in that kind of thing. So, functionally, by posting and interacting with content in Tildes we are tuning the various algorithmic recommendation feeds that we interact with to view us all similarly.

It's just an interesting side effect I noticed and some food for thought about the effectiveness of a link aggregator or discussion forum at surfacing novel, interesting content we might not find otherwise. In part, this could just be an effect of Tildes being kind of small and having lots of self-selection biases for its user population. Perhaps if it was more diverse we'd be exposed to more things that break the mold and recommendation algorithms won't be able to pin it all down as easily. In fact, we may be able to use this effect as a way to test the breadth and diversity of content and types of people a site is attracting.

Hey everyone,

I launched my little project LinkLonk here on Tildes back in December and wanted to tell you how it has been going and get your feedback/suggestions.

New changes since the launch:

• The temporary accounts now automatically get deleted after 30 days of inactivity. I didn't have the deletion logic at the time of the launch, but had it implemented about 30 days after launch. Automatic account deletion is quite destructive - removes the account from the database (thank goodness for foreign keys and cascade deletes) and from Firebase Authentication. I'm happy that there were nobugs when I ran it the first time.
• In addition to submitting external links you can now create text posts. The posts are Markdown-formatted (similar to Tildes). One novel thing is that you can post "anonymously". The database has a record of who the author is so the author can delete/edit their post, it's just the name is not show next to the post.
• Comments - each item has a comment section. The comments are ranked based on how much you trust the people who upvoted each comment (as opposed to being pure popularity). This is the same ranking system that is used to rank the "For you" page, but now applied to comments.
  • Unlike Tildes, the comments have a downvote button. The downvote does not bury the comment for everyone else. Instead, it makes your trust in upvotes of people who upvoted that comment go lower. So the downvote button effects what you see, not what others see. It is much harder to abuse that button that way. For that reason I feel much more comfortable putting it there. However, there is a second order effect. If you downvote a comment that someone else already downvoted - then you will trust the downvotes of that person. When they downvote some other comment - then it will rank lower for you. In a sense they earn your trust to moderate content for you by identifying comments you don't want to see.

In terms of users, there have been 260 user records created (some from my shameless plug comments on HackerNews). Of those, ~45 rated something - excluding those that were temporary accounts and were deleted. And I think we have 2 regularly active users (excluding myself). In my mind I had 10 as the number of active users that I was hoping to get by the end of 2021. At this rate we may reach it.

I was pleasantly surprised that there have been no misbehaving users. I didn't need to remove any content even once. This lead me to constantly postpone the implementation of a content reporting system. I hope it stays this way for a long time.

The whole idea of a trust based recommendation system is based on having someone to trust. Right now it is the RSS feeds that are generating most of the content recommendations for the active users. But ideally it would be mostly users recommending content to users. I have two priorities for the near future:

• Make the "single-player" experience better so the active users find value already. As an example, I added full-text search through items you liked
• Find more users to improve the "multi-player" experience. One option is to submit a "Show HN:" post on HackerNews. But you can only do it once and I'm not sure I'm ready to use that shot yet.

What do you think I should do next on these two fronts?

If you would like to give LinkLonk a try register with code "tildes" at https://linklonk.com/register. Feel free to comment on this post: https://linklonk.com/item/6347369602224750592

I built a link sharing website where you connect to users that share your interests. When you upvote a link - you connect to other users who upvoted that link and LinkLonk shows you what else these users upvoted.

The more in common you have with another user the more prominently their other recommendations appear on your list.

The intuition is that the more useful your past recommendations have been for me, the more I can trust your future recommendations.

This is how trust works in meatspace - we keep track of how positive our experiences have been with other people and use that track record to decide who we can trust in the future.

Except that mechanism does not work online. It just does not scale to the numbers of users we interact with. We can remember around 150 other people (the Dunbar number). Beyond that our builtin trust mechanism breaks down. We revert to more coarse and primitive trust mechanisms such as tribalism and mistrust in everyone.

While we cannot personally keep track of every user on a platform - that is what computers are good at.

That is the idea behind LinkLonk. You don't need to remember the names of users who you can trust (in fact there are no usernames on LinkLonk). You simply upvote content that was useful to you and LinkLonk constantly keeps track of how useful every other user has been and ranks new content accordingly.

Another important part of trust is that if you misplace your trust in someone and they let you down then you need a mechanism to stop trusting them.

This is what the downvote button is used for: when you downvote an item, LinkLonk reduces your “trust” in other users that upvoted it. As a result, you will see less content from those users.

The above describes the basic idea. There are a couple more concepts:

• You start off weakly connected to all users, which means that at first you see content sorted by popularity. Rate something and refresh the page - the ranking will change.
• You are not limited to a single persona/interest. If you have multiple interests then you can create a separate collection for each of your interests. When you upvote a link you can choose what collection it belongs to. For example, if you are interested in woodworking and music then you can create two collections and put woodworking links into one and music links into the other. Then other people who liked your woodworking recommendations will only see your other recommendations from the same collection and will not get your music. This is mostly a way for you to help other users find relevant content. It’s optional. You can put everything into the “default” collection if you don’t feel like organizing.
• LinkLonk has another source of recommendations - RSS feeds. When you upvote a blog post LinkLonk connects to the RSS feed of that blog - as if it was another user. LinkLonk pulls updates from the feed and shows you the new entries using the same ranking algorithm: the more you upvote items from the feed the higher the other items from the feed are ranked. You can submit any RSS url and LinkLonk will connect (subscribe) you to it. My hope is that in the early days when we don't have many users you would find LinkLonk useful as a sort of an RSS reader.
• Moderation. When you downvote an item then you get connected to other users who also downvoted that same item. In other words, you will trust their other downvotes. If they downvote something then that item will rank lower for you.

Give it a try at: https://linklonk.com/register with 'tildes' as the invitation code. The invitation code can be used multiple times and I will keep it active for a few days. After that please DM me to get a fresh code.

I’m posting this on Tildes in part because I like the group of people that Tildes has attracted. And I also feel the topics of trust systems, content curation and moderation are relevant to Tildes and to its users (see: https://docs.tildes.net/future-plans#trustreputation-system-for-moderation).

What do you think?

There was a story posted to Hacker News, The Return of the Super-Elite from Jacobin magazine. It was on the front page for a little bit of time. I refreshed and it was on the 2nd page.

5 hours later and it's down to #113, page 4. It has 88 points. The second youngest submission on page 4 is 16 hours old. On page 3, the youngest item is 6 hours old, and has only 7 points. So this article is newer, has a respectable amount of points but within 5 hours has been relegated to page 4, whereas an item that has fewer points and is 1 hour older is sitting on page 3.

edit: the rank keeps dropping, when I first wrote this post it was at #111, then #112, and when I submitted it was at #113, I just refreshed and it's at #114. Other submissions near the range of points and hours are ranking on page 1. On page 5 all items are from 1, 2 or 3 days ago.

I've noticed that any pro-unionization talk seems to disappear much more quickly than other stories.

So let's get our tinfoil hats on and ask is Hacker News suppressing leftist articles or suppressing articles of a certain type altogether?

Or maybe it's just a conspiracy of a bad algorithm for determining where submissions rank?