blue_berry

This idea combines a neuro-symbolic AI system (take a LLM and use it to generate logical code; then make inferences from it, see Neural | Symbolic Type) with Attempto Controlled English, which is a controlled natural language that looks like English but is formally defined and as powerful as first order logic.

The main benefit is that the result of the transformation from document/natural language to the logical language would be readable by not IT experts, as well as editable. They could check the result, add their own rules and facts, as well as queries.

I created a small prototype to show in which direction it would be going (heavily work in progress though). What do you think of this? Would love to here your opinions :)

The fundamental idea of this paper is for ChatGPT-like apps to lose natural language for less energy consumption and more determinism in their answers based on controlled natural languages like ACE; for the user to be able to modify this trade-off-ratio at will based on LLMs (which is not possible when starting from a ChatGPT-like app); and to capture this new paradigm in a new type of browser that has natural language as its primary interface, here called a semantic web-first browser.

In its finished state, the software is supposed to work like this:

• Users start Anthem, which act as a peer in the network
• Once they have found other peers, they connect to a master node, which coordinates the training
• All peers then train a model with their uploaded songs
• Then they generate songs from their local models
• Each user can now listen to and generate their own AI songs

Would love to hear any feedback from you guys :) Does this make the current AI situation better or worse? I could imagine it acting as a protest against AI companies (similar to what Napster did), but I'm not sure which effect it will have on smaller artists (after all, Napster basically lead to music labels coming under the hood of Apple's ecosystem, which in the end, wasn't so good for artists but on the other hand also brought the Fediverse into being ... I'm not sure).

I did a prototype implementation of a “network of ML networks” - an internet-like protocol for federated learning where nodes can discover, join, and migrate between different learning groups based on performance metrics (Repo: https://github.com/bluebbberry/MyceliumNetServer). It's build on Flower AI.

Want do you think of this? It could be used to build a Napster/BitTorrent-like app on this to collaboratively train and share arbitrary machine learning models with other people while keeping data private and only sharing gradients instead of whole models to save bandwidth. Would this be a good counter-weight for big AI companies or actually make things worse?

Would love to hear your opinion ;)

Von Neumann’s idea of self-replicating automata describes machines that can reproduce themselves given a blueprint and a suitable environment. I’m exploring a concept that tries to apply this idea to AI in a modern context:

• AI agents (or “fungus nodes”) that run on federated servers
• They communicate via ActivityPub (used in Mastodon and the Fediverse)
• Each node can train models locally, then merge or share models with others
• Knowledge and behavior are stored in RDF graphs + code (acting like a blueprint)
• Agents evolve via co-training and mutation, they can switch learning groups and also chose to defederate different parts of the network

This creates something like a digital ecosystem of AI agents, growing across the social web; with nodes being able to freely train their models, which indirectly results in shared models moving across the network in comparison to siloed models of current federated learning.

My question: Is this kind of architecture - blending self-replicating AI agents, federated learning, and social protocols like ActivityPub - feasible and scalable in practice? Or are there fundamental barriers (technical, theoretical, or social) that would limit it?

I started to realize this using an architecture with four micro-services for each node (frontend, backend, knowledge graph using fuseki jena and activitypub-communicator); however, it brings my local laptop to its limits even with 8 nodes.

The question could also be stated differently: how much compute would be necessary to trigger non trivial behaviours that can generate some value to sustain the overall system?

cross-posted from: https://lemmy.world/post/28384872

This is a showcase of combining vibe coding with the Fediverse and attempto controlled english (ace).

I'm fascinated by vibe coding, but I'm also highly critical of it. It fascinates me, because it enables people, who normally cannot code to be able to generate running code. What I don't like, is that it just isn't actual programming. It's closer to a wishing well. It fosters a quasi-magical understanding of programming and computer science, which is already too common in current society (I wrote a paper about it here: https://philpapers.org/rec/BINAKR). That's why, in my opinion, the Fediverse should set a counter-point here with something like a first-order logic language like ACE, which actually brings people closer to an actual understanding of computer science concepts like modeling and logic without hiding the complexity behind seemingly "magic", and could also result in better code.

The above demo shows a glimpse of how this could look like on the Fediverse. Imagine communities being able to form their own spaces on the social web through language! Simply using natural language will probably not be specific enough here. We always imagine everything getting much easier, but that's just the logic of digital capitalism that tries to sell us innovation as inventing yet a more easy way to get your coke handed to you, which can only lead to more and more environmental destruction. So, what will the language interface for the future digital look like? I think it will be more something like the semi-formalic language found in technical manuals, cooking recipes and judicial texts. Something like ace, in between coding, domain specific languages, modeling and natural language. And people who are experts at this and know the old technical stuff that no one understands anymore will be the new "coders". But maybe I'm wrong.

Repo: https://github.com/bluebbberry/AceCoding.social.