Tim ChambersFeb 12, 2023

From: @pluralistic:

"To understand the future of Mastodon, we have to understand network effects...a product or service that gets more valuable as it attracts new users.

The people who stay...make it a busier, more exciting place...the next surge will find an even higher equilibrium as users who try Mastodon find more fully acclimated users who can hold their hands as they get settled in, and also provide the vibrant community that presents a good reason for doing so."

https://doctorow.medium.com/of-course-mastodon-lost-users-c48ef8102891

Of Course Mastodon Lost Users - Cory Doctorow - Medium

This week (Feb 10–17), I’ll be in Australia, touring my book Chokepoint Capitalism with my co-author, Rebecca Giblin. We’re doing a remote event for NZ on Feb 13. Next are Melbourne (Feb 14), Sydney…

Medium
Show threadSteffen Christensen
@tchambers @pluralistic I'm of the opinion that the killer app in the space will be user-controllable algorithmic timelines. We all like the "no algorithms" feature of Mastodon: it is obviously better than the rage-fuelling and democracy-destroying timelines that attracting eyeballs to advertisers requires of Facebook and Twitter. https://www.wsj.com/podcasts/the-journal/the-facebook-files-part-4-the-outrage-algorithm/e619fbb7-43b0-485b-877f-18a98ffa773f
The Facebook Files, Part 4: The Outrage Algorithm - The Journal. - WSJ Podcasts

In the fourth episode of our investigative series based on an extensive array of internal Facebook documents, we explore the fallout of a major algorithm change the company made in 2018. The documents outline how an emphasis on engagement incentivized the spread of divisive, sensational content and misinformation. WSJ's Keach Hagey and Jeff Horwitz explain how attempts from within the company to undo some of the damage were often thwarted.

Wall Street Journal
Feb 12, 2023 at 12:05pmWeb
Show threadSteffen ChristensenFeb 12, 2023

@tchambers @pluralistic However, chronological timelines are also suboptimal:
- I post some great things at 3 in the morning Eastern. No one will read those.
- I want to oversample Cory's posts in my feed. There is no way to do that.
- Some posters post 50 tops per days; others,1 toot per week. The latter tend to be better. I want more of those. Too bad for me.

The problem with algorithmic timelines isn't that they are algorithmic. It's that you, the user, have zero control over them.

Show threadSteffen ChristensenFeb 12, 2023

@tchambers @pluralistic Now that there is a large user base on Mastodon, we can think about how to make algorithmic timelines that actually work for us.

A good solution would be to allow upsampling per-follower of the most popular threads using a Google circles-style interface.

Show threadSteffen ChristensenFeb 12, 2023
@tchambers @pluralistic A great solution would be to use GPT-3's Ada model to encode embeddings of every toot on a server and then upsample based on shared basis vectors. You could downsample toxicity or rage farming at the same time. I don't know the average foot length, but the average tweet length is 34 characters, maybe 17 BPEs. At that price, $1 would buy you encodings for 147,000 toots.
https://openai.com/api/pricing/
Pricing

OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity.

OpenAI
Show threadSteffen ChristensenFeb 12, 2023
@tchambers @pluralistic I have about 200,000 lifetime tweets over at @[email protected]; that would be about $2 in OpenAI costs to encode.
Show threadSteffen ChristensenFeb 12, 2023
@tchambers @pluralistic I've done some experiments on affinity mapping short segments of text for work using the next-larger model, Babbage, and it works very well!
I found only one misfire among my 18,000 task descriptions in all my testing. I suspect that it would work well for algorithmic timelines under your control as well.
Show threadFreebirdFrolicFeb 12, 2023

@Wikisteff @tchambers @pluralistic

So peaceful here in Mastodonia, the spaces gently punctuated with belly gaffaws, enlightening viewpoints, facts and Cats. Also current events.

Show threadJoanna Bryson, blatheringFeb 12, 2023
@Wikisteff @tchambers @pluralistic I've been meaning to write about this for Wired, but haven't found time. Twitter was great when it had no algorithm other than muting tweets that started with an @ if you didn't know at least two people involved. I have to say that I haven't found mastodon as great as twitter was then, at least not yet. Also, Twitter was fine with tapbot & lists, avoiding their apps & algs.
Show threadWim🧮Feb 12, 2023

@Wikisteff
I agree, with growing numbers of users, I think it will become a very useful thing to have.

https://octodon.social/@wim_v12e/109602587133689475

@tchambers @pluralistic

Wim 🅾 (@[email protected])

Content warning: On the effect of fedi growth and algorithms

the Octodon