Replying to
@valkyrie@wandering.shop this looks pretty cool, is there source available anywhere?
Remote
kworker 💻🏴🌱
@kworker@kolektiva.social
mastodon 4.6.8Anarchist | Hacker | Math Appreciator | Music Lover | Nature Enjoyer | Sociable Nerd
I dig anarchist thought ranging from insurrecto to social ecologist, but I like action even better. Expand degrees of freedom, attack what oppresses us, build harmony with the ecosystems we're part of.
b"\x41\x43\x41\x42"
0 Followers
0 Following
10 Posts
Joined December 18, 2022
Pronouns:
he/him
Location:
Occupied Potawatomi/Anishinabek Land
Replying to
@regehr wow, a refreshing look at LLMs that isn't simply marketing hype! Really appreciate it. Interesting to see the kind of work and strict scoping this takes. Also seems like IMO LLMs still really don't outweigh the negatives & cost
Replying to
@regehr yeah I hear ya. I won't repeat a laundry list that I think many people are familiar with, I just think LLMs have turned out to be very negative for society
Are there any #OpenSource #Software licenses that are #AntiAI? Or actually effective efforts to keep the scrapers away? Large corporations, among many other capitalists, have gotten a lot of value from running freely available open source software, and we've put up with it as a kind of bargain because it meant more adoption and maybe some slim opportunity to make some money or get collaboration.
But now this freely available code has become the training set for large models to rube goldberg copy-paste from. If licensing our code with something like the #GPL ever did anything at all, can't we come up with licenses that forbid or imposes significant cost on using the code to train an #LLM or other AI system? Is there any effort with teeth to find sloppily-pasted GPL code in closed-source codebases to GPL the whole thing and yank it from whatever company owns it?
Why?/The Rant: Open Source is, imo, about more than simply being able to read code for software you use. It's about establishing a commons of computation, where people are empowered to use computers as they see fit, not as a corporation deems profitable. Communities and individuals can better their own lives and share regardless of what the Microsofts and Oracles of the world think. But that has to be adversarial to the corporations because they will want to crush or capture that commons, which means it either goes away or becomes a private park that is no longer for you (if you're even allowed in). It also has to be ecologically sustainable. If your commons destroys the world around you and disrupts your ability to grow food, then commons or not it won't exist for much longer (and maybe neither will you). Which basically means that nearly all LLM use simply needs to stop, and open source communities must reject it. We need to, in any way possible, remove the fruits of our labor from their training data.
Replying to
@adamgreenfield@social.coop @fishsauce@mastodon.social amazing, I literally could've found it with a quick search! And here I was expecting it to be boxed up in someone's basement. Excited to dig into it
Replying to
@adamgreenfield I think a lot of "mesh network" attention is given exclusively to meshtastic, but this doesn't do that. I think reticulum is more promising long-term, but needs a much more user-friendly experience.
Replying to
@platypus@glammr.us this one's best with popcorn. No sympathy for someone that essentially pipes the Markov chain text generator into their shell. Sorry but random and unexpected behavior is kind of the deal there, maybe they could spend all that token money on someone who knows not to rm random production data.
Replying to
@adamgreenfield@social.coop ok I read through the full pdf and have thoughts! They don't stray far from my earlier skimming, but they're based on actually reading it now.
I agree with the Meshtastic -> Reticulum progression and I'm pursuing that myself. I really love the good advice throughout the book on organizing and the social aspect of things. The tech is important but organizing with others is what will actually make or break this. It's also broken down into lists, roadmaps, pitches, and useful tips that make it much easier for someone new to organizing. The recommendation to have food at a meeting is an easily underestimated thing that is very important. I also really like that adapting these strategies to your local context is covered with an example of LA. I look forward to examples in other areas, especially rural ones. I also like the emphasis on just getting started with a friend, and that this can help get something off the ground even if there's nothing else in the area yet. I also think the advice on keeping up momentum and avoiding drift is good, as well as treating reticulum as a next step for people going in depth on mesh networking, rather than a requirement for everyone to switch to.
I'm somewhat bugged by a few things like talking about meshtastic/reticulum as "layers", and as dirt roads/highways. I think this can give the false impression that they interoperate and that reticulum runs on top of meshtastic. I also wish the nrf52840 based boards were covered more, imo the limited battery life of esp32 based boards make them mostly useful for grid-based or proof-of-concept things, and serious off-grid meshtastic needs to use something more power efficient.
I also feel like there's some tech entrepreneur mentality seeping in with things like covering "first mover" advantage. It is real that people who are early in a "space" can influence it a lot, and so imo should help establish welcoming norms for communities that push back against capitalist norms like running everything as a business and trying to monetize everything. But to its credit the book does also talk about the mesh as something one person doesn't own or control. I also think there's a missed opportunity in assuming the reader is the first one experimenting with meshtastic in their area, they may find other people already using it that they can connect with. Also meshcore would've been good to at least mention, but I understand omitting it to keep things streamlined. It also didn't mention the typical shortcomings someone would experience in meshtastic, like only getting half of a conversation because of the low hop count.
I think my biggest criticism is the "web 4" framing and the explicit inclusion of "web 3". I do think there's people into "web 3" who are into privacy and autonomy who could be good to bring into community mesh efforts, but IMO they have to be willing to ditch the capitalism obsession that would bring in baggage like unnecessarily cramming Merkle trees (excuse me, I mean blockchains) into things. There was already a LoRa "web 3" company called Helium that IMO is a better example of corporate capture of these kinds of things than an actual community network. "Web 3" is an old marketing buzzword that's most associated with the amazing NFT scam, unregulated securities and a scummy rugpull-of-the-week industry that has some very specific useful applications. Meanwhile reticulum offers a way to build an autonomous data network that can use whatever existing hardware is laying around that feels most promising for building a resilient ability to communicate regardless of governments/ISPs/power grids, and without being locked in to a particular technology or business model. It's more of a departure from the web as it is than a continuation of increasingly desperate business models like the "web 1, 2, 3, 4" framing implies. IMO the "web 3" world is fully tied in with capitalist pipedreams that are killing us like hyperscalar AI data centers, while the best of the mesh network efforts are building a tangible step away from that death cult, which is a huge part of what makes them interesting. Your off-grid autonomous community network isn't actually off-grid or autonomous if you need an LLM in a hyperscalar datacenter to write its software for you, but I digress.
"Web 4" silliness aside, I overall like this book and think it's great for encouraging people to just get started, and providing enough of a practical roadmap to actually follow through on it. Getting started and keeping momentum is the hard part, but once an active community forms you'll get active tech experimentation and opportunities to bring new people in. You'll also have the "web 3" acolytes balanced by other people ;). As much as I'd nitpick or criticize things in the book or the project as it's presented, if this gets people to start a local group of people getting to know each other and building out a mesh network that people actually use, that's a big win. Such things can't really be reduced to a number, but 7/10
Replying to
@adamgreenfield@social.coop wow I wonder if the rulebook is in some archive somewhere, I'd love to read it