João Costa
@JD557@blog.joaocosta.eu
Follow
Portuguese software engineer at Kevel.
My instance is running on a small server so please #nobot
- GitHub
- Itch.io
@adamwarski if one part of a system is not decentralized, then the whole system cannot be considered decentralized. If all the decentralized systems are required to go through a centralized system at any point, then it's a centralized system.
@BeAware Yeah some of its components are. That's probably why BSky more often talks about "credible exit". But for that to be possible, there need to be centralized alternatives as well.
@adamwarski does there, though? Mastodon's migration feature works pretty well and a lot of Fedi devs support importing from the mastodon format.
I'd say Fedi beats Bluesky with that as well because if there's only one option, then it's not an option. Which is currently the case with Bluesky/AT Proto.
@BeAware True, that's why we need at least two *significant* options to start talking about "credible exit". Otherwise there's nowhere to exit to ;)
@adamwarski@softwaremill.social
The "[X] Algorithm" is open-source;
I see this being thrown around a lot, and it's start to annoy me more than it should (not by the way it's done on this article, but how some people use this as a proof of X's transparency).
I had some doubts about calling it open source when it was released, and now that there are no commits in a whole year, I find it even more dubious.
My main concern being with a lot of the features being blackboxes. Taking this to a ridicule, it would be like claiming that def getScore(tweet: Tweet): Future[Double] = callRecommender(tweet).zip(callToxicityDetector(tweet)).map(_ * _) is an open source recommender.
For example:
-
There are some US politics features. They seem to be unused in the source code (maybe the algorithm is adjusted near election time?), and it's not obvious how the labeling required for those features is done.
-
On the topic of US politics, there also seems to be some code that handles those annotations in a closed source timeline mixer
-
There are also a lot of references to an HML model (which I also believe to be closed source) to handle toxicity.
This is all from a quick skim at the code. To me, it kind of feels that this repo is enough to be called open source and claim transparency, while at the same time:
-
It cannot be reproduced/tested locally (too many closed-source parts)
-
It's too big to be auditable without a huge investment
-
It doesn't stop anyone from "adjusting the knobs" as they see fit.
Just to be clear, I'm not claiming that anyone is using this in a malicious way. My complaint is specifically about the "Open Source" part. I feel like "partly open-source" wouldn't be as misleading.
@JD557 Thanks for the write-up, this does put a different perspective on things. I agree that "partly open-source" is a much better description - I'll use it in the article, if you don't mind.
Is #Bluesky decentralized? How does it compare to #Mastodon and #X?
A technical overview of the current state of the network, and how it might shape up the future:
https://softwaremill.com/blueskys-decentralized-architecture-compared-to-mastodon-and-twitter-x/