Flushing Meta and heading for clearer skies (Newsletter January 2025)

I hope you had a good end of year break — well, at least better than mine, which I spent half of underground sorting out a sewage leak.

(This is the January 2025 edition of my newsletter. Browse them all, and subscribe.)

Luckily, before that bountiful splendour gushed forth from its overloaded drain, I found the time to articulate how I’d apply what I’ve learnt so far about Bluesky and the ATmosphere to myhub.ai. Currently I’m showing it to some friends and integrating into my ever-evolving experimental wiki, and hope to have more later this year. And as some of the links below show, I’m not alone. 

And while on the subject of uncontrolled torrents of sewage: if the spectacle of recent days hasn't convinced you yet, please consider leaving Meta and/or X. I left all things Meta on Monday, and posted a how-to thread of screenshots to Bluesky. On X, I'm still maintaining the holding pattern I adopted last year, when I still hoped that the nazi running the place would one day be stripped of it by investors fed up with his losses. That hope is looking increasingly forlorn, but given that a Trump-Musk bustup is almost inevitable, I’ll give it another couple of months.

I hope you enjoy the resources I think are worth reading below.

Mathew

---

The Best Ofs

So far, there are only 3 resources on my Hub tagged the Best Stuff I Like about Everything, This year, but then the year's just begun.

January 6 and the Justification Machine

The study of mis- and disinformation is not new (see the 250 resources on my Hub, but given the date I couldn't not mention January 6 and the Triumph of the Justification Machine, which uses the June 6 Capitol Hill riots as a prism to investigate misinformation.

It provides many examples of how the internet works as a "justification machine... providing denial-as-a-service... in real time" as the riots unfolded. The end result is a nihilistic culture "where every ... human success or tragedy becomes little more than evidence to score political points" and the "breakneck pace of our information environment... [produces] a feeling of being trapped in an eternal present tense."

I Highlighted this article because of those last 3 words. We need to slow down.

19 things we learnt about LLMs

Then there's Simon Willison's "review of things we figured out about LLMs in the past twelve months, plus my attempt at identifying key themes and pivotal moments". It has 19 major points, all worth reading, and all worth transforming into things to check out in greater detail.

For me, the standout point was that "Synthetic training data works great", so model collapse is "clearly not happening ... AI labs increasingly train on synthetic content ... The days of just grabbing a full scrape of the web and indiscriminately dumping it into a training run are long gone". This has obvious implications for one of the business models I hope could help power decentralised collective intelligence, but I still think that authentically human, high-value content created by communities could become valuable to future AI development, for example to provide seeds for synthetic training data.

Bluesky and the ATmosphere gathers steam

The third "best" resource this year was the announcement of Free Our Feeds, which is raising funds for an independent foundation to support the development of the ATmosphere and build "an entire ecosystem of interconnected apps and different companies that have people’s interests at heart". It aims to launch at the end of this year.

So what will they fund? One idea doing the rounds is a second Bluesky relay* - as Cory Doctorow put it in Billionaire-proofing the internet, "if a body independent of the Bluesky corporation implements its federation services, then we ensure that its fire exits are beyond the control of its VCs".

(* What's a relay? If you need one more intro to how Bluesky actually works, I recently added How Does BlueSky Work? to my collection of Bluesky guides, and it's one of the better ones).

But a lot of people - myself included - would love to see grants and seed capital to support new product launches. Because every new product brings in new users not just to that product, but the ATmosphere as a whole, increasing the userbase for the next product.

For the same reason, I hope at least some of those grants will go to bridging the ATmosphere with the Fediverse. It's therefore encouraging to see that one of FreeOurFeed's custodians is Mallory Knodel, exec director of the Social Web Foundation, who suggests a range of ideas in Free Our Feeds and Algorithmic Pluralism.

The wrong question: "Is it decentralised?"

Because Fediverse vs. ATmosphere is not an either/or question, and "is it decentralised?" does not have a Yes/No answer - there's a spectrum of decentralisation (and plenty of other dimensions besides). Each protocol already does its version of decentralisation well, and there's room for applications on both.

From my perspective, the heat and friction was taken out of the Fediverse vs. ATmosphere debate by the exchange of blog posts by Christine Lemmer-Webber & Bryan Newbold, summarised in my last edition. I really liked the leadership shown by both.

Since then, CWL has responded with "one more post", while I found a mathematical analysis of the ATProto's scaling issue also useful. As I see it, hopefully today's increasingly-healthy social protocol competition will evolve until the two spaces complement each other, as:

• ATProto's shared-heap model structurally favours few servers, "and thus has a natural centralizing force", as CWL argued
• ActivityPub has an incentive to keep the average followers/user low.

Hence the Fediverse will probably remain a good place for cozyweb spaces, and co-exist with more global conversations on the more centralised and business-like ATmosphere. The world needs both.

The public sector's slow scroll towards decentralisation

So the world - in the shape of governments - needs to get on board. Last timeI checked, however, for every 7 Bluesky or Mastodon accounts run by the EU institutions, there are 930 on platforms owned by Musk and Zuckerberg.

Of course, I'm being unfair: they've been investing in Meta and Facebook for 15 years, so the comparison is invidious. At least their share tool now includes Bluesky and Mastodon (as long as you scrollllllll..... - wait, Netvibes???).

Anyway, over the break I developed some ideas for how the online communities convened by the public sector, notably to support public participation in policy, could benefit from integrating with Bluesky and the ATmosphere. I hope to have something to publish later this year.

Science of non celebrity

In the meantime, scientists are increasingly on board Bluesky, although Science's analysis gives only two cheers: "Bluesky fosters collegial interactions among scientists, but potentially limits interactions beyond the academic community... [and] is unlikely to ever become as huge... There’s not going to be a new Twitter. There are going to be a lot of different things."

As I pointed out above, that's a good thing in general. In science, moreover, Twitter's "popularity contest drove academics toward what social media users valued... [but] sells short academic expertise”. Hopefully science on Bluesky will be less celebrity-focused.

Nature, not to be outdone, surveyed their readers, getting almost 6000 responses, 85% of which said that they were from working scientists. Just over half had left Twitter, and used Bluesky mainly “to connect with other scientists, keep up to date with other research or researchers, and promote their own research”. In their qualitative comments the common adjectives were “more pleasant, more supportive, friendlier, kinder, nicer, more collegial, uplifting, more peaceful and safer … debate there is more measured and more focused… fewer Nazis”, although also “some find Bluesky boring … a leftwing echo chamber … full of woke crazy people”.

You can’t please all the people, all the time. Nor should you try.