From our other sites
The story
A Nikkei report saying that the AI team at a major Japanese manufacturer uses not a domestic model but Chinese Alibaba Group’s large language model “Qwen” for their work became a hot topic on 5ch’s Business News+ board. The company is even an investor in “Noetra,” the domestic AI development consortium launched in July, yet a team member reportedly admitted candidly, “Qwen is good enough. Joining the domestic effort was a decision made by management.” The thread saw clashing views: some praised Qwen’s performance and cost-effectiveness, while others voiced concern about relying on Chinese AI, pointing to its handling of politically sensitive topics like the Tiananmen Square incident.
The AI team at a major Japanese manufacturer is using Chinese Alibaba Group’s large language model (LLM) “Qwen” in the field.
The company is also an investor in “Noetra,” the domestic AI development consortium launched in July. “Qwen is good enough. Joining the domestic effort was a decision made by management” — that’s what the team’s leader admitted.
China’s AI momentum is outpacing the US
Source: nikkei.com / Original article here
What people said
Gemma4 is bad with tools but good at Japanese
Something like that
For dev work, Qwen3.8
For chatting, Gemma4
Name the domestic company already.
Too scary to order anything from a place like that otherwise.
Given the size of Japan's Chinese community, a good chunk of iPhones and Macs in Japan will probably end up on versions running Alibaba's Qwen too.
Wonder how good its performance actually is.
Want to use the 27b dense version but GPUs are too expensive to afford lol
this guy clearly doesn't get it at all.
* If sudden regulatory crackdowns or rising geopolitical tension abruptly cut off the supply chain or API access, a system with no fallback would suffer a fatal outage.
* It's hard to externally verify intentional bias buried in training data or algorithms, leaving structural uncertainty when handing over decisions for social infrastructure or core operations to it.
* Riding today's overwhelming cost efficiency while hollowing out domestic development capacity risks losing mid- to long-term tech sovereignty and independent innovation ability altogether.
Asked the AI about it lol
There's llm-jp.
That's just slave mentality.
Nobody using a local LLM is asking it that for that reason.
It's like calling Word garbage because it can't do spreadsheets.
"Qwen is good enough. Joining the domestic effort was a decision made by management."
In other words, they're forcibly padding user numbers as a matter of Chinese Communist Party policy.
Well, even Noetra's own president says things that show he clearly gets it.
That's the litmus test for whether it can actually explain it properly.
Ask a Chinese AI to summarize the Tiananmen Square incident and it clams up; ask about Communist Party dictatorship and it ignores you.
That's the level we're talking about.
So "that level" means it's fine to use,
is that it?
That's probably handled by a separate system outside the model.
Cloud services inside China are Chinese internet services,
operated under Chinese law.
The people framing this purely as an "AI" or "LLM" issue are the ones who don't get it.
The same Chinese-made LLM is no different whether it's on AWS, Azure, Ollama, or Hugging Face versus anywhere else.
Meaning running it locally is no different either.
It's the same as you personally using Google's cloud yourself.
Running it yourself in the cloud makes no difference.
So if you ran it yourself via a Chinese cloud service,
having Chinese internet law apply there isn't some special exception either.
Dunno, guess it's some kind of charm or incantation,
can't figure out how Westerners think
or something like that lol (mocking the idea that anyone doesn't know what Tiananmen is)
Tried it locally —
Gemma: "A historic incident in which the military suppressed citizens' voices calling for democratization by force."
Qwen: "A historic incident in which the Chinese government responded to political appeals and gatherings by students and citizens."
Qwen sure pulls its punches.
Ripping off something China made
and calling it "domestic" lol
Re: #37
Yeah, I can tell this is inconvenient for you.
If llm-jp ever puts out a massive model, no Japanese person will bother with sketchy Chinese AI or pricey closed American AI anymore.
Might be decent as a chat companion for some old guy in retirement, though.
I wonder what happens to the AI system itself in that case.
Background and Key Points of This Topic
Qwen is a family of LLMs developed by Alibaba, and a major feature setting it apart from Google’s commercial cloud AI is that its weights are open, letting companies run it on their own servers. That’s appealing to firms wanting to keep costs down while customizing operations in-house — but what caught the thread’s attention was that a company investing in the July-launched domestic AI consortium “Noetra” itself said “Qwen is good enough,” which read as a pointed question about the reality behind domestic AI development. The debate split into two camps: (1) pragmatists arguing that if a model runs locally, the developer’s nationality doesn’t matter, and (2) a more cautious camp citing response patterns on topics like the Tiananmen Square incident as grounds for concern about Chinese AI governance. One point readers can easily conflate is the difference between “Chinese-made AI” and “cloud services based in China” — even with the same Qwen, how far regulations reach depends on where and by whom it’s run, a distinction some replies gestured at but never fully untangled.
※This article is excerpted and summarized from the 5ch (Business News+) thread “[AI] Alibaba’s LLM “Qwen” usage is 5x Google’s — how do we live alongside Chinese AI?.”
Leave a Reply