AI and the Fediverse: How Can We Do More than Just Be Haters of AI

Convener: Anca (@anca@mastodon.xyz)

Notes

• Anca: How can we make AI our own, similar to how Linux is the OS response to proprietary operating system. How do we “build the open source version”?

• Kjetil: How can we strengthen the collective without choking the individual. Can we create loyal agents? Personalizing your part of the social web that’s good for you?

  • Anca: Do you use some technology that you’re looking at?

    • Kjetil: Working on a new project in start of September, but nothing to share yet.

• Jeremiah: Existing between 2 extremes: husband working at large company on AI product who feels empowered by it and my online community which is extremely opposed to it. Somewhere in the middle: finding it useful sometimes, error prone often, but unquestionably empowering.

  • I’m exploring running local models. I have a huge computer. And don’t have the environmental, privacy, and ethical concerns because of running it locally.

  • Olares PC: https://www.olares.com/

  • Amazing Christine Lemmer-Webber post that’s critical: https://dustycloud.org/blog/faulty-towers-vibe-sickness-and-the-vibe-bobsled/

  • This is the best perspective on this moment in AI I’ve read. https://blog.ceejbot.com/posts/future-shock/

• Anca: building a product that is using local AI models

• Georgia: Glad to have the session because talking about it on the fediverse is not welcomed. We need citizen AI coders to help with development of the features we know are inhibiting the UX quality of social web services. Now we don’t need loads of money and developers to build useful tools. Main problem is adoption and that’s from quality of social web products/services.

• Unique opportunity w/ Fediverse and AI - ActivityPub means that content doesn’t sit in just one place. AI can be used to show the actual content from Fedi, rather than regurgitating the most likely sentence.

  • Jeremiah: I think using AI to surface information rather than summarizing it is underrated.

• Aditya: Self-identified AI hater. However, best to not police how people use AI. Better to scrutinize how institutions use it, how governments implement it. Environmental challenges, concentration of power, commons like Wikipedia struggling on amount of traffic but not getting much financial support back. How open are open models? Not having understanding of dataset the models are used to train on. Labor angle: training requirements immense amount of labor being cheaply extracted.

• Scott: Bombarded by people any time saying anything not negative about AI. Blast radius of OpenAI makes having nuanced discussion difficult. Confused implementation with the technology. There are horrible companies, but also ways to talk about it and use it.

  • It’s slightly shameful to use Claude, but my own local model needs a much too big machine to run on. I want to be entirely local in 2 years.

  • Apertus OpenModel: https://apertus-ai.org/

  • Empowering as a designer to not have to argue with a developer to make progress

  • Hoping in 2 years to have powerful local models

    • Jeremiah: I agree. Apple has effectively infinite money and it’s not building enormous datacenters right now like all of its competitors. I think it’s betting that local models get faster and better sooner than there would be a return-on-investment in that kind of capital expenditure.

• Samir: Claude is so good at coding now, working as well as a senior engineer. We need to understand it and how to use it. The people most resistant to it will be left out. Not an AI champion, but impressed by progress and finding it useful. Never needing to hire engineers or designers in the future?

  • Anca: I’m not sure I agree that we won’t need to hire these people - but what is an engineer or a designer in the age of AI?

• Bjorn: It solves a lot of problems that were already solved, but the people didn’t know about the solution. Using AI has other sideeffects, but the option of learning how to just use the tool is not being considered an option. Lots of people will lose jobs or be reduced to “the human in the loop”. Proportionality of when to use AI or not.

• Georgia: The idea of not being a programmer and being able to make solutions for the problems they have is now part of the zeiteist. Social apps are something people are focusing on. If we don’t properly harness this technology, we will miss an opportunity. Want these people who are willing to take things into their own hands to join us instead of reinventing the wheel. Contribute and build without having to fall into legal ptis.

  • Anca: They are making the technology, without the community

  • Bjorn: Empowerment provided by AI is one side of the coin. The other side is dependence on those able to provide you with the models (training data etc) anca How do we get the empowerment without the dependence?

  • Scott: The early days of mobile were much the same way. 100s of goofy little apps that completely disappeared. It’s something we have to go through

  • Anca: Who has tried to train their own model? It’s an educational experience that might help limit the dependence aspect.

  • we need better documentation for ActivityPub so vibe coders can spin up compatible solutions

• Kjetil: Not afraid of AI, but the men who control AI.

  • We should have started 20 years ago with a societal conversation around the Commons. Now, companies are extracting from the Commons, with little giving back to grow the Commons. We need to focus on creating virtuous cycles, and monetization cannot be the only way to do that, because often, it isn’t.

• 2 immediate use cases for AI in Fedi that don’t need to be LLM’s:

  • Search

  • Content Recommendations vs Chronological feeds (not working well for him, moved to Bluesky for this reason, built own algorithmic feed)

  • Challenge of people not opting-in to be discoverable

    • Jeremiah: https://www.fediscovery.org/ project for discovery and search

• Matthias: In Open Source, we often see AI as “the bad guy”, but I think that it wil stay, and we need to be part of that discussion, to take that discussion and find sustainable alternatives and solutions.

  • Good use cases for people who are disabled, or people who can’t speak a language, who use AI to help them have conversations.

  • WordPress has a lot of AI going on. Curious if I should also be using it.

  • Anca: Went to WordPress US camp: people using AI to help make their installations more secure, making useful plugins

• Jeremiah: Recent discussion of wild idea to “just rewrite Mastodon in Rust as a drop-in replacement with a WordPress-style plugin architecture” so communities can customize it to meet their needs and add features that Mastodon has not been able to deliver yet. Would it be vilified for just existing, immediately defederated, and doomed from the start to never be able to have a chance to benefit people? Probably and that’s unfortunate.

  • Anil said we should not be afraid of the existing user base if we are aiming for the next billion users.

• AI useful, but needs lots of hand holding still. People probably putting too much faith in the machine at the moment.

  • Humbling experience to have LLM look through my code and find all my bugs. Had it pull in all the FEPs and examine how I implemented them wrong.