i wonder what you guys think about the open weight chinese AIs that are directly competing with the big tech AIs, isnt it a good idea to support projects that have open source and open weight only to make them more advanced than the closed ones from greedy billionaires? i say that because the AI bubble is a huge threat for the elites currently, and if we support their competition it might finally pop.
and the outcome will be a decentralized thing rather than they succeeding on replacing jobs and ruining our lives more than they already do. what yall think about this idea? remember that this is an economic war in the end, and the only way to take down the ones we despise is by decentralizing everything they want to centralize.
im asking this because i noticed people here in lemmy.zip are (justifiably) anti-ai… but only begging the govs that are directly lobbied by the elites wont work. we cannot bet on the government to bite the hand that feeds them.
It’s not open-source, and it likely has the same censorship embedded in the model.
Pros:
Most of them are open weight and self hostable.
Its cool to see a non USA company competitior .
Cons:
Even open weight AI (LLMS) have some ethics problems (i.e, training on everything that is a copyright violation)well yeah true, but the difference is that it wont be owned by the elites and they wont have more power than they could have if the best tools were theirs and only theirs.
Suffer not the abominable intelligence to live.
They Shall Not suffer a Machine to Think! For Ruin Shall be its Purpose and Accursed be the Work.
oh?
I think it really depends on what you’re concerned with. Open-weight models provide you with a bit more control. For example you won’t leak all your private information to some mega-corporation. But they still centralize power, have a big impact on the environment, labour market… They also hallucinate and flood the internet with misinformation, bots and made-up stuff… It’ll also still be tuned to fit someone’s agenda. Whether that’s the bias and morals Mark Zuckerberg, Elon Musk, Sam Altman likes to push down on the world. Or a Chinese “startup” attached to some Chinese government sponsored tech company. You pick your poison…
I don’t think there’s a noteworthy chance this will end up as some decentralized thing. It’ll always be researched, trained and designed by whoever is able to afford those kinds of salaries and datacenters. Which is going to be the elites, billionaires, largest companies and governments.
thats why we should focus on copyleft licensing everything we can and find a way to create copyleft idea patents (only permits the idea usage if all of the projects linked to such idea will also be copyleft-patented and will only use copyleft licenses for code,art and the likes)
i could say im even more copyleft extreme than even richard stallman… i am completely against the concept of copyright and knowledge ownership. every method and knowledge must be publicly available to be used by everyone.
we do have laws to make this happen, we need just to create a big enough copyleft scene to spread the copyleft environment and make it so its expensive and less optimal to do things in a proprietary way.
I think money is the major factor which does the gate-keeping. Let’s say I’m not okay with the other (commercial) models out there. What they do and don’t do, their tone and political bias. Like Elon Musk claims… Now I’m gonna need some 6-digit sum of $$$ to train my own model. And a couple of thousand wage-slaves in a poor country to curate datasets for me, do RLFH. And that’s the real kicker. Musk can do it easily. But I wouldn’t know where to get that kind of money. And it’s prohibitively expensive for community projects. And even large independent organizations like universities struggle to do AI research on the same level as OpenAI, Anthropic, X, Meta, the Chinese, … do it.
I think even if we changed copyright, piled up large, state-of-the-art public datasets, forced them to release the weights, we’d still be in a similar situation as of today. Where we get some breadcrumbs tossed by someone. We can choose whose breadcrumbs we pick. And we can put some topping on it. But it’s not really emancipating in the same way copyleft works for software.
The whole approach the Chinese are taking to AI is entirely different from the way the feudal techlords of the USA are going at it.
The Chinese approach is one of pragmatism: AI projects have stated goals and desired ends. They do actual engineering, chief among of which’s approach is testing. If the AI project shows measurable (important word there!) progress toward the stated goal, it is permitted to continue. If it doesn’t, it is killed rather brutally. And it’s not (necessarily) the state that does the decision-making, though it often is.
For a solid example, my Honor (a one-time brand of Huawei, spun off into its own independent company now) phone has a baked-in translator that I’d become quite reliant upon for the more difficult aspects of some of my applications. Then, one update, it was replaced with an LLMbecile chatbot that was an utter fucking nightmare. The old system was “select text, copy it, press the little widget that showed up at the end of my screen” and I’d get a translation, plus some abilities to switch things like source and target language, specific dialects, etc. Then, one day, I pressed the little widget and I got a degenerative AI chatbot. One that was so stupidly configured that it would translate selected Chinese text into Chinese. Never mind that the phone was configured for English. So there was an added step each time: type out “now translate this to English” or some such.
I stopped using the translator after that, and apparently so did a lot of other people because two updates later, the old translator was back. The experiment had been done. The stated goals were not met. The project of using an LLM for the translation engine was abandoned. Compare and contrast to how AI is infiltrating every app made by the American techlord feudalists. AI is injected. People hate it and refuse to use it. The technofeudal assholes double down and force it even harder.
So while the ethical issues of LLMbeciles remain with Chinese AI engines (it’s all based on stolen input), and while the environmental (and fiscal) costs are much lower than equivalently capable western tech—something that may mitigate some people’s concerns—at the very least the Chinese approach to incorporating AI into things is very pragmatic and not jammed down your throat harder if it makes you gag.
I still avoid LLMbeciles, even those made here in China, but I think of the evils of degenerative AI, the Chinese ones are the least evil.
But still evil.
hah thats exacly how i see it. i just want to minimize long term damage as much as possible
My opinion is the name of the comm: “Fuck AI.”
That said, I do recognise that the enemy of my enemy is my friend, but that’s a flawed perspective. The enemy of your enemy might see why you are your enemy’s enemy, especially if they want the same things while being in opposition. Thusly, the opposition to AI could be an enemy to both western and eastern AIs, even if those AIs themselves are enemies to one another.
I think the prevailing logic with regards to supporting any Chinese tech is "I don’t care what Beijing knows about me because I’m less likely to go somewhere where they have jurisdiction than somewhere Washington has jurisdiction (which is basically anywhere). If you’re someone the Chinese would never care about (like my middle aged working class ass), there’s something to be said about that.
But I don’t hate AI because the Epstein class supports it, I hate it because I think of it like Skynet from the Terminator franchise, or the machines from the Matrix movies. The truth is, AI is older than those concepts and we probably all use some form of it. It’s just all the AI slop we see online that forces us to take a position on it one way or another. I don’t remember hating AI or other people really talking about hating AI until a couple years ago. Stealing art, deepfakes, and driving up the costs of computer hardware, largely. Before that, most of us didn’t care nearly as much. My wife’s an artist, and I used to build computers — component pricing is a big part of why I’m a Mac user now. I don’t regret my choice, but I probably had another build or two left in me. My Macs will last past my 50th birthday for sure, so safe to say I’m “out” of building. I did always want a Mac since I was a kid, but also, component pricing pushed me toward that, so it’s part of why I say “fuck AI”. It became personal to me.
my distaste for AI isnt due to “x group vs y group” but more cuz i am aware the goal is to centralize power through automation, but i talked specifically about chinese AI because they do have a push for decentralized ones (open weight)… if AI be decentralized they cant gather all the power for themselves (no matter the group) and the elites do a takeover cuz they dont need the population anymore.
if one day china stop supporting open weight, then we go support whoever else does, because open source and open weight and anything copyleft is what takes power away from whos on the top.
Its pretty shit and its not open.
how is deepseek and kimi not open? also lets not forget that the training costs are absurdly lower and requires less hardware power, this alone is much more environment-friendly than whatever openAI and google and others are doing.
Im sorry but how is having less power costs related to open source? Also they arent more environmentally friendly because they were trained in china which means it was likely trained off a coal power plant.
But actually to the point, its just open souce washing by releasing the model weights. Nothing about it is open source.
It has restrictions on use. It isnt available to be studied or modified because they have not released the data or pipelines and any derivatives inhert the restriction on use.
Its a proprietary model and its released because there is no point in not releasing it since it trails behind frontier US models. Meta released the blueprint on how to train new models off frontier models for a fraction of the cost years before the first chinese model. Dont get me wrong they’re interesting and China is a big player in the AI space but thats it.
