capitalist world requires huge subsidies to make their AIs kinda work, and those models are expensive, bloatware, resource intensive and mostly crappy anyways
socialist world makes open source models, mostly free to use or very cheap, lightweight, easy to configure and customize, and pretty decent
how socialism always is better playing the capitalist game than capitalist class themselves?
The worst part is OpenAI and the likes made good non-LLM AI in their early days, they just stopped because these models were made to get people acquainted with AI and get them social recognition. Like the Whisper speech-to-text (transcription) model is honestly still pretty good, and people keep building on it to make it more efficient or qualitative. It came out in September 2022!
They have the talent and the resources to pool this talent together and make stuff, they just misapply it.
I’m glad to see people are flocking to Chinese AI models. The west is going to have a real hard time with its usual monopolistic practices if people can just go to cheaper Chinese models and I’m here for it. :)
And just now Anthropic has been forced by the US government to stop providing Fable and Mythos to foreign nationals (ie non US citizens) because they’re scared of the possible uses of it. Translation: people will get used to other LLMs and never touch Anthropic again. (The model is probably not even that much better than others at adverse capabilities like hacking and bypassing security)
That sounds like a real shoot themselves in the foot moment, if I’m reading it right. And yeah, I’m doubtful that these US models are actually much better. It’s possible some of them have an edge for the time being because of getting there first with the right hardware for it, but China is much better equipped to catch up and surpass them in the long-term since its industry and education is not kneecapped by late stage capitalism constantly.
Yeah that’s an absolutely incredible self own by the burger reich. No company outside the US is going to want to rely on American models going forward if they can just be yanked from under you without notice.
Yep. I’d say we’re entering a new phase of technology and at its height, the empire probably would have been prioritizing proliferation of its tech as the solution for everybody, for the soft power of it. But the empire’s soft power is in decline like the rest of it, Trump keeps tearing up soft power like ribbon with his overt mobster ways, and so China can just step in and say, “What if you didn’t have to deal with all that bullshit? All the nasty strings attached and the rugpulls?”
Honestly, the soft power is evaporating faster than I expected. I guess we really are hitting the whole quantity transforming into quality moment here.
I think vassal states will likely still be forced to buy, because they exist to feed the beast, but yes it’s likely as the EU maybe stops being a lapdop that American software will continue to fall out of favor to ensure state independence and industry security.
Does anyone use deepseek for programming, what does your workflow look like?
The easiest way it to use opencode, you get an API key from DeepSeek and put it in your shell environment. Then fire up opencode in the project folder and ask the model to give you a phased plan for the feature you want it to implement, review it, then tell it to build it one deliverable at a time, on a branch, review the work at each step, and go to the next. Doing things in phased fashion is the easiest way to make sure things stay on track and you understand what the code is actually doing.
I don’t use deepseek, but I do use qwen3.
I mostly use this: https://github.com/madox2/vim-ai
I open a file, and ask AI to write something. Then I start adjusting it, or fixing the mistakes.
if i feel a need to keep up llm use once google enshittifies i either won’t or i’ll use deepseek, simple as
I find deepseek is straight up better for search than google at this point, and it also does a way better job including non western sources in the results
I haven’t used it, i haven’t gone out of my way to use llm shit aside from seeing google insert it into search results and watching the quality go from where it was giving people poison cake recipes to now where it’s generally pretty good at generating a plausible answer to whatever, and I haven’t seen it fuck up food at all
Do lightweight models such as DeepSeek still require a server room filled with GPUs to run?
I’m speaking in the context of AI-assisted coding. A loop of feedbacks, hooks, and self-reflections to reduce errors and bugs generated by AIs.
This kind of automation is somewhat effective time wise but extremely costly in computation, memory, hardware, and energy.
I guess as long as we rely on GPUs, NPUs, TPUs, or any kind of silicon chips, the dream of democratizing (the means of production of) AI is still far fetched.
You can run Qwen 3.6 27b on a laptop with less than 32gb ram now, and it does a surprisingly decent job at coding tasks. There are also tools like this which dramatically improve local model capability https://github.com/itigges22/ATLAS
