Four independent Chinese companies released extremely good open source models in the past few months (DeepSeek, Qwen/Alibaba, Kimi/Moonshot, GLM/Z.ai). No American or European companies are doing that, including titans like Meta. What gives?
I like Qwen 235 quite a bit too, and I generally agree with your sentiment, but this was a very large American open source model.
Unless we're getting into the complications on what "open source" model actually means, in which case I have no clue if these are just open weight or what.
The Chinese are doing it because they don't have access to enough of the latest GPUs to run their own models. Americans aren't doing this because they need to recoup the cost of their massive GPU investments.
Why is inference less attainable when it technically requires less GPU processing to run? Kimi has a chat app on their page using K2 so they must have figured out inference to some extent.
Inference is usually less gpu-compute heavy, but much more gpu-vram heavy pound-for-pound compared to training. General rule of thumb is that you need 20x more vram for training a model with X params, than for inference for that same size model. So assuming batch size b, then serving more than 20*b users would tilt vram use on the side of inference.
This isn't really accurate; it's an extremely rough rule of thumb and ignores a lot of stuff. But it's important to point out that inference is quickly adding to costs for all AI companies. Deepseek claims that they used $5.6mil to train Deepseek R1; that's about 10-20 trillion tokens at their current pricing- or 1 million users sending just 100 requests at full context size.
That's super wrong. A lot of why people flipped out about Deepseek V3 is because of how cheap and how fast their GPUaaS model is.
There is so much misinformation both on HN, and in this very thread about LLMs and GPUs and cloud and it's exhausting trying to call it out all the time - especially when it's happening from folks who are considered "respected" in the field.
If they were doing that I expect someone would have found evidence of it. Everything I've seen so far has lead me to believe that these Chinese AI labs are training their own models from scratch.
Just one example: if you know the training data used for a model you can prompt it in a way that can expose whether or not that training data was used.
You either don't know which training data was used for say chatgpt oss, or training data can be included into some open dataset like pile or similar. I think this test is very unreliable, and even if someone come to such conclusion, not clear what is the value of such conclusion, and if that someone can be trusted.
My intuition tells me it is vanishingly unlikely that any of the major AI labs - including the Chinese ones - have fine-tuned someone else's model and claimed that they trained it from scratch and got away with it.
Maybe I'm wrong about that, but I've never heard any of the AI training experts (and they're a talkative bunch) raise that as a suspicion.
There have been allegations of distillation - where models are partially trained on output from other models, eg using OpenAI models to generate training data for DeepSeek. That's not the same as starting with open model weights and training on those - until recently (gpt-oss) OpenAI didn't release their model weights.
> An unnamed OpenAI executive is quoted in a letter to the committee, claiming that an internal review found that “DeepSeek employees circumvented guardrails in OpenAI’s models to extract reasoning outputs, which can be used in a technique known as ‘distillation’ to accelerate the development of advanced model reasoning capabilities at a lower cost.”
Additionally, it would be interesting to know if there is dynamics in opposite directions, US corps (oai, xai) can now incorporate Chinese models into their core models as one/several expert towers.
At ECAI conference last week there was a panel discussion and someone had a great quote, "in Europe we are in the golden age of AI regulation, while the US and China are in the actual golden age of AI".
"Who could've predicted?" as a sarcastic response to someone's stupid actions leading to entirely predictable consequences is probably as old as sarcasm itself.
>Moonshot 1: GPT-4 Parity (2027)
>Objective: 100B parameter model matching GPT-4 benchmarks, proving European technical viability
This feels like a joke... Parity with a 2024 model in 2027? The Chinese didn't wait, they just did it.
The timeline for #1 LLM is also so far into the future that it is entirely plausible that by 2031, nobody uses transformer based LLMs as we know them today anymore. For reference: The attention paper is only 8 years old. Some wild new architecture could come out in that time that makes catching up meaningless.
Honestly, do we need to? If the Chinese release SOTA open source models, why should we invest a ton just to have another one? We can just use theirs, that's the beauty of open source.
For the vast majority, they're not "open source" they're "open weights". They don't release the training data or training code / configs.
It's kind of like releasing a 3d scene rendered to a JPG vs actually providing someone with the assets.
You can still use it, and it's possible to fine-tune it, but it's not really the same. There's tremendous soft power in deciding LLM alignment and material emphasis. As these things become more incorporated into education, for instance, the ability to frame "we don't talk about ba sing se" issues are going to be tremendously powerful.
* Our satellites are giving us by far the best understanding of our universe, capturing one third of the visible sky in incredible detail - just check out this mission update video if you want your mind blown: https://www.youtube.com/watch?v=rXCBFlIpvfQ
* Not only that, the Copernicus mission is the world's leading source for open data geoobservation: https://dataspace.copernicus.eu/
* We've given the world mRNA vaccines to solve the Covid crisis and GLP-1 antagonists to solve the obesity crisis.
* CERN and is figuring out questions about the fundamental nature of the universe, with the LHC being by far the largest particle accelerator in the world, an engineering precision feat that couldn't have been accomplished anywhere else.
Pioneering, innovation and drive forward isn't just about the latest tech fad. It's about fundamental research on how our universe works. Everyone else is downstream of us.
I’m confused. Who is this “We”? Do you realize how behind in many respects most of Europe is? How it’s been parceled up and destroyed by the EU? Science projects led by a few countries doesn’t cut it.
It’s not propaganda at all. The standards of living there are shit. But enjoy the particle collider, I guess?
We is Europe. Like everywhere else, we are behind in some aspects and ahead in others.
> The standards of living there are shit.
Now you're just trolling. I've lived in both the US and in multiple EU countries. Let me tell you, the standard of living in the US does not hold a candle to the one in the EU.
The answer is simply that no one would pay to use them for a number of reasons including privacy. They have to give them away and put up some semblance of openness. No option really.
I know first hand companies paying them. Chinese internal software market is gigantic. Full of companies and startups that have barely made into a single publication in the west.
Of course they are paying them. That’s not my point. My point is this is the only way for them to gain market share and they need Western users to train future models. They have to give them away. I’d be shocked if compute costs are not heavily subsidized by CCP.
But the CCP only has access to the US market because they joined the WTO, but when they joined the WTO they signed a treaty that they wouldn't do things like that.
I don’t think there’s any privacy that OpenAI or Anthropic are giving you that DeepSeek isn’t giving you. ChatGPT usage logs were held by court order at one point.
It’s true that DeepSeek won’t give you reliable info on Tiananmen Square but I would argue that’s a very rare use case in practice. Most people will be writing boilerplate code or summarizing mundane emails.
Love their nonsense excuse they they are trying to protect us from misuse of "superintelligence".
>“We believe the benefits of superintelligence should be shared with the world as broadly as possible. That said, superintelligence will raise novel safety concerns. We’ll need to be rigorous about mitigating these risks and careful about what we choose to open source.” -Mark Zuckerberg
Meta has shown us daily that they have no interest in protecting anything but their profits. They certainly don't intend to protect people from the harm their technology may do.
They just know that saying "this is profitable enough for us to keep it proprietary and restrict it to our own paid ecosystem" will make the enthusiasts running local Llama models mad at them.
Also, the Meta AI 'team' is currently retooling so they can put something together with a handful of Zuck-picked experts making $100m+ each rather than hundreds making ~$1m each.
Too bad those experts are not worth their 300 million packages. I've seen the google scholars of the confirmed crazy comp hires and it's not Yann Lecun tier that's for sure.
