Nvidia, Microsoft, Meta warn against overregulating open-weight models

I don't know that HN usually does. But they did buy a decent amount of publicity by (initially) telling the US government no in using the models for war.

1. Breaking Points Discussion: https://www.youtube.com/watch?v=SIUrshGwgDI

2. Full Bloomberg interview: https://www.youtube.com/watch?v=x2VHFgyawPE

Same experience with Anthropic's. I applied for my employer, and.. 0 response.

(Grade 3 AI which can hack both previous tiers is exclusively sold to the highest bidder.)

This is wrong IMO. You should have a serious conversation about your products security with someone who is actually trained on that subject. LLMs are useless if you don't already know more about the thing than the LLM, or if you don't care too much about the outcome (internal tools etc.)

And as a more general point - more major competitors in a domain is very good for everybody except those competitors themselves, who would rather there be as few as possible.

So Nvidia benefits from more inference providers running many Chinese models that they otherwise would not have been able to service.

Microsoft and Meta are also-rans at this point. Their best hope of catching up is probably leveraging their infra and open models.

If I were cynical, I'd say it isn't helped by the current US administration.

Nvidia sells the hardware.

Microsoft sells the hosting.

It's also the case that it makes it hard to attract customers if your openweight model is banned. A major reason the likes of Qwen, Kimi, GLM, and Deepseek are popular (well, at least highly talked about) is because of the open weight models they gave away.

Could you elaborate? I don't understand your implication.

Amazon in particular hosts open weight models via Amazon Bedrock (e.g. Ministral 14B 3.0, DeepSeek V3.2, ...)[1].

In case it changes the context of this comment, I am a Meta employee, all opinions are my own.

[1] https://docs.aws.amazon.com/bedrock/latest/userguide/model-c...

If we go down the line of dead internet theory which I’m becoming more convinced of these days, the volume of information that’s not necessarily original or extracted from reality and just interpolated and extrapolated from existing information in different ways by LLMs will greatly outnumber human information coming up.

In which case these models should.. start to converge on the same data I imagine, with slightly different behaviors within that. One big generative orgy feedback loop.

No wonder openai and anthropic are terrified

The gap is shrinking, but considering their models still lag our frontier, I wouldn't say they "run circles" around us.

[0] https://en.wikipedia.org/wiki/Export_of_cryptography_from_th...

Therefore it is no surprise that both companies are among the signatories.

The motivation of Palantir is very transparent, they have launched a product created in cooperation with NVIDIA, which is a turnkey system (Sovereign AI) that includes all hardware (i.e. a rack with servers full of NVIDIA GPUs) and all software needed by a company to run on its premises LLM inference and also training/fine tuning.

Thus the Palantir product competes with OpenAI and Anthropic and it benefits from the existence of open weights LLMs.

The motivation of Microsoft is more obscure for now, but because its criticism was very similar to that of Palantir, about how dangerous it is for corporations to delegate the processing of their AI needs to external entities like OpenAI and Anthropic, I assume that Microsoft is also going to propose an alternative to using remotely the OpenAI and Anthropic APIs.

Unlike Palantir, Microsoft is not likely to suggest self-hosting on premises, but I suppose that they might promote self-hosting on some kind of instances rented from Azure.

(a) sellers of hardware (Nvidia, IBM) and those who rent hardware out to others (Amazon, Microsoft)

They want open models because they don't have to license it to run it on their hardware, so they can offer lower cost and inspect what they are offering to clients

(b) open model creators (Arcee, Perplexity, Mistral, IBM, Microsoft)

...who would be unable to work if open-weight models are banned

(c) Software-heavy companies ( Mozilla, The Linux Foundation , ServiceNow, Microsoft, IBM, heck all of them)

...who want models that are cheap to run because that reduces the cost of running an LLM (whether you are trying to replace a software developer or just assist them, the value of a reduced cost LLM is directionally the same)

AI models don’t really have nationality. They don’t have race/ethnicity fields on their model cards. It’s impossible to determine the country of origin of a model by looking at their weights. There is no DNA test for AIs. They exist in a mathematical space where human-made borders make no sense.

This means IF there is a regulation for open weight models, it has to apply to ALL open weight models. There is no any other option, since you can always train/fine-tune a model to change it’s weights and rebrand it as a new model.

MS is playing literally all the sides in this whole AI thing. They are a massive investor in oAI, and have lots of rights to use their models for a long time. They also have the cloud where they offer inference w/ all the benefits of already being there, data retention, etc. They also have a research wing that can build models (rumours are they're aiming for frontier-ish). I don't think there's currently an angle in this AI thing that MS hasn't aimed for.

OpenAI and Anthropic can bring American open-weight model providers like Meta to court but not the Chinese. Chinese companies breaking American IP law (whether the law is correct or not) are basically immune. This has been going on forever, I guess enough is at stake finally for the federal government to care.

If I change one weight in Kimi, is it still a Chinese model?

If I fine-tune Kimi on pro-America nationalistic freedom loving anti-Chinese propaganda, is it still a Chinese model?

If I distill Kimi from one server to the next without ever directly transferring any weights, is it still a Chinese model?

If Claude accidentally trains on some of Kimi's output, is Claude then considered to be a Chinese model?

If China's next open-weight model is released secretly through a European company, is it still a Chinese model?

Is the US planning to invade China and take away their models? Because otherwise China is still going to have the models even if the US bans them. I don't think they're going to be going with the US position on this one.

there's nothing open source about an open weight model. It's more like the binary blobs Linux fought so hard against. If you can't compile, or in this case train, from source then you're missing the "source" part of open source.

https://news.ycombinator.com/newsguidelines.html

And besides, it seems to me like the ethical choice. Given how these models are, in some very real sense, mechanical plagiators, built on the generosity of creators past and present, some of them now in danger of being replaced by the machine. I think the least the labs can do is open these models up. These and other such considerations were the reason OpenAI started with that name. Of course, it was questionable that those ideals would survive the encounter with generational wealth. Just look up what the founders of Google were saying about advertising when they were two students tinkering at an as of yet unproven tech. Same thing for OpenAI, self-interest speaks that much louder when there's real money on the table.

It just boggles the mind that people now make excuses for their all-too-predictable about-turn.

Instead a bunch of tech companies are gathering to try to stop OpenAI and Anthropic fear-bouncing the White House and the Republican Congress into giving them regulatory capture and repeating the mistakes they are making around RISC-V.

Those mistakes won't just entrench two companies, they will entrench the bigger-better-faster-more model (closed companies making ever bigger cloud-bound models) when it is abundantly clear that enormous progress can still be made on smaller, even desktop-bound models (where, due to distribution, open weights are essentially inevitable).

Regulatory capture that stops open weights work will also have impacts on local and on-device AI work, as well as on academic research.

It would seem as if the community either isn't doing that or is relying on the Chinese to do that.