I have all the same issues most of y’all have with the moral and environmental issues with giant corporate models but I take issue with this statement:
The core function of an LLM is to enclose public commons (aggregate, open-access human knowledge) in a centrally-controlled black box.
The core function of an LLM is to generate statically plausible text (which is what my totally open source mobile keyboard is doing as I type using a very small transformer based LLM, for instance)
Using it to provide an answer to a search instead of returning sources is 100% the evil you described. But it is a shitty use for a technology that would be unfair to reflect entirely on the technlogy itself.
LLMs are not going away. We might disagree on their usefulness (I flip flop daily on my opinion about it, which is usually a sign that something is inherently neutral) but zealot blanket rejections worry me a bit.
The other knee-jerk reaction about energy (and water, but that is not unavoidable) usage is also something I try to process a bit compartimentalized. It needs to improve and the scale of growth is unsustainable. Does that invalidate everything currently explored or researched?
The push for more efficiency is vital and rightful. Do more with less. But while it’s fair to criticize someone for using an incandescent light bulb instead of better technology to, say, illuminate a room, criticizing them for using light in that room is wrong, IMHO. We don’t need less light (well, yes, outdoors, but for different reasons), we need better technology and cleaner energy so we don’t need to worry about who is turning on which light. I get that “AI” is power hungry, and that needs to improve, but I am very uncomfortable with the idea that we should decide a priori if something is worth using energy or not. It’s… A bit draconian?
I know its not a super original position (“a tool is just a tool”). I’m trying to work through this myself. As I type this I think of PoW blockchains as a counterexample that I would bring up to debate myself. Yes, it looks like there are usages that appear to be inherently “wrong”. Why do I find blockchain worse? Because I consider it unworthy of the energy spent for it, which makes me “guilty” of what I criticized…
Damn, It’s hard to try to have opinions!
More in topic: vibe coding (super icky name, jfc) might be vaguely OK for prototyping in some cases, or extremely limited cases where you can almost prove correctness. Or yeah, personal tools where you’re the only person to be responsible and affected by the results. Anything more than that, and it makes me nervous. It has not much to share with solarpunk per se. But AI aided development (maybe a broader and less silly named concept) is not antithetical to solarpunk, IMHO. The DIY nature you ( @strongoose@slrpnk.net) describe doesn’t go down at infinity. To build a community garden from scratch you first need to invent the universe. You not knowing how to invent the universe. You still own the technology if you use a tool you don’t fully understand the internals of. You need to retain the option to understand it though, I agree.
Clearly a mistake to focus con electrification where we don’t control the supply chain. Better to keep pumping oil where we… * checks notes *… Yeah, well, end of the press conference, nothing to see here.
I don’t say that biofuels don’t have a reason to exist but I have a feeling they are mentioned just to say something “alternative” to electrification