Got it, thanks for the explanation.

That's an interesting approach. Mastodon instances could show the federated feed rather than local only though, am I wrong? I'll have to read more on this 😊

I'd like to learn about the two approaches to existing online of the two protocols. I thought they were slight variations on the same concept of self-sovereignity and independence from big tech, and the surveillance economy. Do you have a primer to educate myself on their different visions?

That's my fear too: Protocols may not be *the* solution if the problems are somewhere else. I gave up on Mastodon because of the frustrating experience with the whole federation. I'm using Bsky (and Threads), more polished but I wonder if more ATP servers will make it more like Mastodon. We'll see.

I think I understand what you mean. My experience with Mastodon has been frustrating, that's why I decided to pull the plug and just use my blog as ActivityPub server.

Well, that's my point: Creating a new server (Eurosky) would remove that advantage, right? One has to pick which one to use to signup, am I wrong?

10/ If we want to preserve both the creative rights of content producers *and* the open development of AI, we must recognize that licensing-only regimes will fail. The answer lies not in private ordering, but in more robust solutions to govern data access. www.promarket.org/2025/11/19/t...

9/ The practical impossibility of scaling licensing does not resolve larger normative questions like fair use. But it decisively undermines the claim that licensing can realistically replace fair use or open-scraping for internet-scale data ingestion.

8/ For smaller AI entrants or #OpenSource developers, licensing-based training data access would likely remain out of reach. The result: fewer competitors, higher barriers to entry, and reduced experimentation — contrary to the spirit of a healthy tech ecosystem.

7/ Over time, this dynamic risks concentrating power not only among AI developers, but also among a narrow set of content gatekeepers — large publishers and platforms. This undermines the diversity and openness of the internet that facilitated early AI innovation.

6/ Any attempt to funnel AI training through exclusive or semi-exclusive licensing deals effectively erects data-moats. That means only deep-pocketed firms can build or maintain competitive models, undermining competition.

5/ Even if we attempted a collective-licensing or blanket-licensing model (analogous to music-rights bodies), it would almost certainly fail for AI training data. We cannot reliably trace which texts, images or works matter for model performance: tying compensation to actual use becomes impossible.

4/ The marginal value of any one individual work — a blog post, a tweet, a niche article — is effectively negligible. The costs of negotiating, clearing rights, and administering payment for each would vastly exceed the benefit. That alone makes mass licensing unscalable.

3/ Most of the internet’s content is held by myriad individuals, small publishers, or user-generated sources — often without any collective rights-management structure. Licensing each of those works individually is simply unworkable.

2/ What those deals demonstrate is not that licensing is scalable — but that it works only on a “big-media” scale, with a handful of large, concentrated content owners.