I think it can very well be applied to the Threadiverse.

• Sin #1: The First-Move Problem - Doesn’t really applies for the Threadiverse, because the instances (at least for me) do feel genuinely different, with a different culture, etc., which is one of the most exciting things for me here
• Sin #2: Navigation Inconsistency - Basically the same here.
• Sin #3: Remote Interaction Hell - Also the same here, right?
• Sin #4: Private Mentions Aren’t Really DM’s - Same here, right?
• Sin #5: The Phantom Social Graph - There definitely are synchronization issues on Lemmy, too (see the australian instance, which, I think has a latency of a week of so per post :D). Otherwise because the Threadiverse is still rather small, it seems to work mostly fine.
• Sin #6: The Discovery Problem - Much better on Lemmy. The algorithms are both transparent and make the threadiverse feel alive even though it has much less user.
• Sin #7: Basically doesn’t apply here, because you don’t follow users, but can be applied for communities. And multiple of the same communities on different instances are a big problem of the Threadiverse. Also abandoned communities. PyFed solves this with topics and Lemmy also has an upcoming feature for this in v1.0 I think.

I think the most pressing issue is sin#7 if applied to communities.

In an abstract sense, I see the Threadiverse as inversion of Mastodon: instead of posting messages to a personal account, which tags may be interesting to you to discover other similar content, in the Threadiverse, users post to hashtags and who posted them is only secondary important to you, but may be used to discover more content by the same account.

Cool. Well, the feedback until now was rather lukewarm. But that’s fine, I’m now going more in a P2P-direction. It would be cool to have a way for everybody to participate in the training of big AI models in case HuggingFace enshittifies

Yeah thats a good point. Also given that nodes could be fairly far apart from one another, this could become a serious problem.

Currently the nodes only recommend music (and are not really good at it tbh). But theoretically, it could be all kinds of machine learning problems (then again, there is the issue with scaling and quality of the training results).

Thanks :)

Good point

It was just a demo. But when I develop it further, it will be either a client or a whole instance-configurator (hopefully).

Isn’t NodeBB compatible with the Fediverse by now?

What do you think of this? I’m all ears for your thoughts :)

Awesome!

I think a link between your idea and mine can be found in the work of writer Evgeny Morozov (https://mondediplo.com/2024/08/07ai-cold-war), who did some interesting research of alternative forms of how the internet could have developed including a project by the chilean government called “Cybersyn” (in his podcast “The Santiago Boys”, https://open.spotify.com/show/7xlRxnooUnl48JVo726YXn). Although it was pretty centralized and not exactly Amazon, more like a socialist distribution system between industries. Well, its a very interesting podcast anyways …

I made a first prototype here: https://github.com/bluebbberry/MyceliumWebServer. Its recommends songs to the users. You can see it here: https://techhub.social/@myceliumweb and try it out by posting to #babyfungus on Mastodon.

You can do AI in an ethical way by making it more decentralized. The idea behind the mycelial web is to realize it based on volunteer computing, meaning that everybody can contribute computing power. And then I can say, for example: use my models, which was trained with all these other models, on this Amazon alternative to recommend me stuff. And the AI model was trained on my PC and runs on my PC (just wasn’t trained solely with my computing power or my data alone).

Could this be an app realized based on the mycelial web - amazon uses huge AI models to predict their customer’s behaviour and do the logistics … (not sure myself, but it probably won’t work solely on ActivityPub)

The idea is that every fungi-node also has a UI, yes. So you would be able to browse the AI models - for example if you chat with bot A, and the bot is currently learning with bot B and C, those bots would be visible to you and you could open their UI, too. And it should also show bots with which it trained earlier, too.

This way you could “browse” the resulting AI web via the browser.

Well, its similar to a botnet, but one that is open and transparent. You can browse the different nodes, etc. And you can (at least hopefully in the future) add your own computing resources to the network to participate in the AI training.

Its a bit like a more sensible version of bitcoin: instead of waisting energy for the proof of work, its training a shared AI model.

(Its running on a Hetzner server.)

Well, big tech has big computers, the Fediverse doesn’t, but we have many small one. That’s why its a good idea to combine them.

This has already been done for example by projects like SETI@home or FOLDING@home.

My idea is to build a web based on this idea.

There is a good video on federated, decentralized AI training here: https://www.cl.cam.ac.uk/seminars/wednesday/video/20241127-1500-t221089.html

One can imagine this project as a decentralized huggingface, which (spoiler alert), could also enshittify.

The idea is that we need bots like this because huggingface can enshittify

Hi, the bot was incorrectly set up and posted every 60s, which was too much and it got suspended.

Its now updated to posting every 12 hours and hopefully, it will be soon up again

No, not yet …

So its basically just a mastodon bot, a knowledge graph and an AI-model, who all work together.

The mastodon bot makes the functionality available to users (they can ask for song recommendations), the AI model is obviously trained and the knowledge graph is used to save the model and for collaborative communication between AI agents.

I’m not 100% sure whether it will be counter-productive or not, but maybe AI on the Fediverse could be a good thing. Like it could push the Fediverse forward (all big social networks nowadays have their own llms, ours should be federated of course). I think combining these three aspects (social web, semantic web/knowledge graphs and autonoumos agents) could be a cool thing. I have noticed how the narratives in the three departments are similar to one another (social web enthusiasts speak about walled gardens, data/knowledge enthusiasts about data silos and AI enthusiasts about big, centralized AI).

The following three resources point to how similar the approaches are:

• https://www.cl.cam.ac.uk/seminars/wednesday/video/20241127-1500-t221089.html
• https://about.flipboard.com/inside-flipboard/dot-social-evan-prodromou/
• https://www.youtube.com/watch?v=Fy5eKqM8-OA

But how to combine them? Maybe as a fungus?