#asknostr #nostr #wheretheboysat?

nostr:nevent1qywhwumn8ghj7mn0wd68ytndw46xjmnewaskcmr9wshxxmmd9uqsuamnwvaz7tmwdaejumr0dshszymhwden5te0danxvcmgv95kutnsw43z7qpq3v5cqgy7tuuu9797dv3tavdmzr3j4yz53s2w5g8vhl2tlcak9z0shawjpf

Reply to this note

Please Login to reply.

Discussion

“Better” is subjective thus what matters is having different algorithms that are easily consumable via a standardized interface (DVMs)

I’m working on baking this in into Highlighter and showing to the other clients what is possible.

Discovery is still broken when there is ZERO reason for it to be broken.

Discoverability should be mind-numbingly easy in an open protocol.

I'm thinking along the lines of a customizable algorithm, I suppose. A personal trending list.

But the base algorithm can also be improved, I think. It's a bit too "dumb".

There is no “base” algorithm though; just too few implementations that everybody uses because they don’t realize there’s better stuff available.

What we need to work on is disintermediating that conception where everybody integrates with one or two different algorithms just because

Oh, I agree, but nostr:nprofile1qyvhwumn8ghj7un9d3shjtnndehhyapwwdhkx6tpdshszrnhwden5te0dehhxtnvdakz7qgnwaehxw309ahkvenrdpskjm3wwp6kytcqyqe4dhnpkwty0ycuazepgzet4wphuzqscrh4zka7jt0qyjqypw9a6vxwe7g was challenging us to improve his specifically.

Then we'd at least have two very different result outputs. The ones we have (band/primal) are so similar in the output, that I mistakingly thought that one was perhaps a fork of the other.

To be honest, it's embarrassingly easy to game the current trending lists, if you are good at pattern-recognition.

If I want something unusual, but interesting, to trend, I just write "GM" or "GN" after the text and then the GM/GN replies push it up the list. 😂

The reason is that there is too much stuff to work on.

Discovery is the main reason people seem to churn. Worth prioritizing it.

Doesn't the slowness of loading from relays also play a role?

What slowness?

1️⃣ Loading of quoted events.

2️⃣ Loading of profile names.

3️⃣ Loading of comments

( Loading of media.

-> Hmm, this could be my problem of using Tor. But somehow people will attribute this to the client if they load slow, or?)

Also how does the user know if all data was loaded for his thread?

I know that we can only check connected relays, so we don't know if really we have all the data. But it would be nice to know if my post was loaded fully from my relays or it is still searching? Maybe it would be a better UI.

This sounds doable, some hint in the UI for that kind of thing wouldn't be bad.

But I think Tor is uniquely slow so most people are probably not suffering as much as you with these things.

I guess more feedback never hurts. But maybe this is personal taste. 🤷

Centralized platforms are slow to load stuff too many times. Nostr threads generally load basically instantly to me, but sometimes there may be some slowness here and there.

In other words: you may be right, but I don't see it as a big problem.

Good argument, you win.

😂

Yes, but this is one of those dull, defining things, like getting replies right, that makes the entire protocol more valuable.

💯💯

One thing that would help a bunch would be collapsing each npub's entries down to their most-popular entry.

Otherwise, we regularly end up with the top-10 including the same person/persons over and over.

Yesterday, for instance started off with:

Jack

Calle

Jack

Calle

Jack

That would also discourage people from large accounts from merely littering their feed with noise, as popular noise would drown their signal off the list.

At the same time, it would encourage other people to raise the signal of their notes, as they have a chance of getting visibility for them.

And it would increase the value of trending lists in discovering new npubs.

This is typical. We sometimes have something like

Odell

Odell

Odell

Gigi

Carla

Gigi

Odell

Gigi

Odell

Carla

Jack

😬

More emphasis on depth of replies and less on length of thread.

Normalized weighting of followers, to capture the Interesting Middle.

Shorten the standard time frame from 24 hours to 12 hours, to keep the list showing where humans are actually located.

Or weighting with a human activity heat map.

Favoring notes that spawn other notes, showing human interest in a topic, rather than mere fandom.

The challenge that places like Twitter seem to have solved is associating posts topically without requiring tags.

Ideally, when you read a note from an algorithmic feed, the next notes you want to see are the ones responding to the topic without directly referencing the original post. Much social media conversation is related to an idea that floats around without a single center. You want to pull all the related posts together to show the larger conversation.

A topical map, with bubbles.

Weeding out zapvertising and hellthreads.

Allowing someone to suppress certain words or their mutes, would be nice.

That can be an easy filter tacked to the end of any algorithm.

Another thing is social-distance. Interactions that happen over a wider social distance (lower WoT between the two) are probably more generally-interesting than those that happen between frens.

This probably reduces the weight away from large accounts, because if everyone follows Jack, then the social distance from any npub to Jack is always low.

Yeah, which makes sense.

It's a feature, even.

A trendlist full of Jack's posts is basically useless. 😂

Everyone already sees those!