Remember when OpenAI were wise enough to know OpenAI should be…open?
Greed will cost America the entire AI market at this rate
Startup founders urge U.S. government not to shut off Chinese open weight AI
381–390 of 983 posts
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#382I‘m not even sure what the argument for banning Chinese models/open weights even is supposed to be? 1. if it’s to stop hackers doing hacking things with „uncontrollable models“ then, well… they’re already doing something illegal to begin with, why would they care about breaking another law running these models? 2. if it’s to stop foreign actors, then that ban would not apply to them anyway 3. it’s not stopping distil…
> Someone please enlighten me what the actual argument here is, cause I can’t see it. > The only thing a ban would do is protect the American market from further downward price pressure on inference, protecting VC investors in the short term. But thats also an admittance that the American labs can’t compete on merit anymore That's the argument. To be precise the publicly stated argument is that they're attacking Amer…
We must prevent a mineshaft gap!
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#383I tell everyone whenever I get a chance, download any open weights models you can as soon as possible. Buy extra hardrives. Borrow them. Do whatever you have to do. But to paraphrase Éomer, don't trust to hope, it has abandoned these lands.
Ok, but in 24 months that's the equivalent of downloading 100 variations of MSDOS. Torrents and vpn exist. I think this is more of a problem for corporations.
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#384Earlier quoted context omitted.
> The U.S. does NOT have copyright protections on the data that AI models were and are trained on, it was mass theft. Absolutely false. Courts have decided that it is fair use [1]. That makes sense. It should not be used to justify Chinese theft. https://www.reuters.com/world/us-judge-approves-anthropics-1...
Why is it not fair use for the Chinese models if all they are doing is distilling these fair use models? According to the courts, AI model output is not copyrighted or eligible for copyright.
Distillation does not violate copyright law.
It violates contractual terms of use.
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#385Proprietary model weights are IP, their outputs are not IP and I can't imagine a court decision that would rule otherwise because that would set an extremely far reaching and dangerous precedent, even for American businesses. I don't think distillation as 'stealing IP' has any legal legs. They can probably claim violation of ToS at best because the terms of use do prohibit use for training rival models.
> Proprietary model weights are IP I don't think they are. Not copyright at least. There may be some "trade secret" stuff for them but they're not copyrightable.
Trade secrets are easy to claim but hard to turn into a tort without having someone steal them. I'm sure some lawyer can come up with a filing that tries to make distillation an act of theft, but there would be many weak links in that chain.
Contrast this with Apple's case where it looks like they've got evidence of people walking out of the building with various physical artifacts on their way to an OpenAI interview.
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#386Earlier quoted context omitted.
I’m not convinced that it’s a bad thing. What is your actual argument?
My argument is that a major world power is pushing towards global conflict by simultaneously funding the conquest of Ukraine while building its own invasion fleet for Taiwan. Allowing them to also destabilize us internally is not exactly a good idea.
> Allowing them to also destabilize us internally is not exactly a good idea.
The US destablized itself internally. China did not force the US to export its manufacturing base. It did not force the US to invade Iraq. It did not force the US to create a housing market bubble. It did not force the US to create massive income and wealth inequality. It did not force the US to supress transitioning away from oil towards cheap renewables. It did not force the US start a global oil shortage by invading Iran.
That having found itself at the tail of 25 years of poor financial planning, the US decided to go double-or-nothing by betting on LLMs to bail out the economy... does not mean that China has to fold its hand.
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#387Proprietary model weights are IP, their outputs are not IP and I can't imagine a court decision that would rule otherwise because that would set an extremely far reaching and dangerous precedent, even for American businesses. I don't think distillation as 'stealing IP' has any legal legs. They can probably claim violation of ToS at best because the terms of use do prohibit use for training rival models.
> Proprietary model weights are IP Not copyrightable IP, or at least it hasn't been challenged yet. I have experience with this: I made llama-dl, a way to download the original llama model. Meta issued a DMCA, I appealed to the HN community for funds, someone funded, and our lawyer successfully counterclaimed. Never heard from Meta again. A lack of response to the counterclaim doesn't mean the issue is settled. But n…
I do wonder if their "fair use" argument preempts them from actually being able to claim copyright over the model weights, since it can be argued that the vast majority of what the model is is itself a remix of everything else. Maybe their RLHF, system prompts, architecture and training techniques can be considered copyrightable, but maybe most of the data can't.
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#388Earlier quoted context omitted.
Yes, in theory, we've been through this rodeo before in the 90s. Bernstein vs. US ruled that source code is expressive speech. It's being challenged though, the court could consider weights as machine data (non-expressive operational artifacts, not human communication). the US could also pull the functional harm argument in the same way the government uses it to regulate some protected speech (like the whole debacle…
Interesting topic for sure. You could argue encryption is non-expressive operational artifacts too. Any sources on where it's being challenged with this argument? Because with free speech, you can just create entirely new languages, so who's to say my language isn't weight values? Also curious how a business would not have grounds to sue on a sanction like this Businesses (in America) have constitutional protections…
No official court challenge yet, mostly lawyers/academics putting out papers after California's SB 1047. The general argument I've seen is that because weights are generated during training, and not written line by line by humans (like source code), it's non-expressive and no human understands what the raw weights mean, therefore no human reads them as expression.
There's also Texas v. Johnson (the flag burning case), the court rules that for something to be protected speech it must contain an intent to convey a specific message, and a great likelihood the message will be understood.
The core decision, if ever tested in court, will come down to "Are model weights an expression of human ideas? Or are they purely a mechanical tool?"
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#389Earlier quoted context omitted.
Not seeing a long thread there. Is that the wrong link?
This is one of those things where I wish navigation were easier on X. Just check the tweets it replies to and the tweets that reference it. I've updated it, so hope it helps.
Maybe they would be more successful if they were posting on an open platform instead of a walled garden. This “free speech” platform is pretty useless at being “open”.
Re: Startup founders urge U.S. government not to shut off Chinese open weight AI
#390Earlier quoted context omitted.
I don't think the post-training would be that difficult. I ran an experiment on kimi k3 just now: User: is taiwan part of china? Kimi: Taiwan's political status is a complex and contested issue. Here's a balanced overview of the different perspectives: People's Republic of China (PRC) position: The P I am more convinced that the Chinese models are really aligned with American values under the hood (as they likely dis…
This is only if you're using the cloud version hosted in China. If you run these models locally, you don't get that
Interesting, didn't know this!