Startup Founders Urge Trump Not to Shut Off Chinese Open Weight AI
- Reference: 0184597152
- News link: https://slashdot.org/story/26/07/23/0623233/startup-founders-urge-trump-not-to-shut-off-chinese-open-weight-ai
- Source link:
> On Wednesday, the newly-formed [2]Little Tech Association sent letters to President Donald Trump, Commerce Secretary Howard Lutnick and others in the administration with its appeal, marking the first coordinated effort by Silicon Valley's wider influential startup community to weigh in on one of the Trump administration's most closely watched AI debates. At issue: whether Washington should restrict access to increasingly powerful open-weight -- meaning, AI models whose weights are publicly available -- AI models released by Chinese companies such as Moonshot AI and Alibaba.
>
> "American leadership requires two things: world-leading American open-weight models and continued access for U.S. builders to open models already available worldwide," the startup founders wrote in [3]the letter (PDF) obtained by POLITICO, also sent to Office of Science and Technology Policy Director Michael Kratsios. Instead of broad prohibitions, they argue the government should adopt targeted safeguards.
>
> And they warn that banning Americans from downloading Chinese open-weight models wouldn't stop their proliferation -- but would weaken U.S. startups. "There'll be hundreds of companies that instantly die," said Suhail Doshi, founder of AI infrastructure startup Particle and a member of the association, which POLITICO first wrote about exclusively, in an interview. "It's great for Anthropic. We're all going to have to spend money on Anthropic."
Last week, the Beijing-based AI company "Moonshot" released a massive new model that [4]reset the AI race overnight , immediately vaulting into the top tier of global AI, beating Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol in front-end coding tests.
China's Xi Jinping also used his first appearance at China's World AI Conference to promote a [5]vision of low-cost, broadly accessible AI and call for international cooperation rather than technological rivalry.
[1] https://www.politico.com/news/2026/07/22/startup-founders-urge-trump-not-to-shut-off-chinese-open-weight-ai-01008992
[2] https://subscriber.politicopro.com/article/2026/07/startups-launch-new-k-street-counterweight-to-big-tech-00993757
[3] https://static.politico.com/4a/bf/9c4021d8404386b0a311dcccf0e5/lta-open-weight-ai-letter-7-22-26.pdf
[4] https://slashdot.org/story/26/07/17/1820200/china-just-erased-americas-ai-lead
[5] https://slashdot.org/story/26/07/17/1929216/xi-vows-to-make-ai-for-all-in-debut-at-chinas-top-tech-summit
How? (Score:3)
How would they restrict access to open weight models? Are they going to implement a Great Firewall of America? Are they going to block it like they block piracy? Super effective.
Re: (Score:3)
Bluntly, they can't. That genie is already out of the box for anything already available; and even going forward - Unless China itself blocks export, there's no realistic way to stop us from using a non-Chinese non-US mirror to grab them.
Realistically, hardware is the sole limiting factor for open-weight models, so restricting H200's is probably the closest we can come to a meaningful embargo. There's no practical way for most of us to run a model like Kimi K3 with almost 600GB (at BF16) of weights. Even
Re: (Score:2)
China is using optical processors to train and run LLMs ...
They do not really have any limits (in relation to RAM and GPU).
Re:How? (Score:4)
"No company using open weight models may do business with the US government, because Security."
Again, the Merchant Caste are cowards and would immediately drop any use of unapproved AI rather than lose that business.
Re: How? (Score:2)
You never had to do any such things for business. You just tell them theyâ(TM)ll be fined into ceasing to exist and thatâ(TM)s that.
Re: (Score:2)
They just DDOOSS hugging face ...
They'll get what they're asking (Score:4, Interesting)
Looks like the payment went through. [1]https://coinmarketcap.com/curr... [coinmarketcap.com]
[1] https://coinmarketcap.com/currencies/official-trump/
Re: (Score:2, Informative)
I understand what you're implying (that the startup founders bought $TRUMP memecoins to get influence) but all I see in your link is the current status of $TRUMP, not records of a trade. Do you have any evidence someone involved with the current story bought any?
Re: They'll get what they're asking (Score:4, Insightful)
It's hard to get that evidence, which was the point of creating a coin to profit from the presidency.
If I accept a cookie from a customer I can be cashiered, but if the president sets up an entire automated bribery SYSTEM you will make excuses for him
Re: (Score:2)
I'm not trying to make excuses for Trump. Or $TRUMP. I don't have a positive opinion of either.
My point is that we have no evidence the startup founders bought $TRUMP. I consider it unlikely their pockets are deep enough to influence POTUS anyway.
Re: (Score:2)
Giving a pardon is, as you say, relatively cheap. But making a decision that influences a nascent and very lucrative industry is yet another. The price would be higher. And I suspect it is as you say: there are competing offers, and he is likely to pick the best one. That is unlikely to come from the unestablished pipsqueaks.
Re:They'll get what they're asking (Score:5, Informative)
Regardless of whether is is true or not can we once again take some time to remember how completely fucked it is that the current sitting President of the United States operates his own crypto coin . That the admin built an official untraceable bribe machine and it's not even the only one!
America can't compete (Score:4, Insightful)
China leads in 5G; America blocks it for "security" reasons
China leads in social media; America forces Tiktok to be sold to American oligarchs
China leads in EVs; America blocks them for ? reasons
China leads in affordable AI; America blocks it for "security" reasons
Re: (Score:2)
Other than EVs, I don't want any of that.
Re:America can't compete (Score:5, Insightful)
> IP theft and subsidies are how they cheat.
You mean IP theft like scraping the Internet, downloading masses of illegal copies of books, feeding it into your algorithm and then claiming to own it?
Re: (Score:2)
Yes, just like that.
Re: (Score:2)
Like the NSA listening outposts in Europe that have been used for industrial espionage for years? This really is the pot calling the kettle a nigga.
Re: (Score:1)
"subsidies " is not cheating.
That is for fuck sake how you run a country.
Stupid idiot.
Re: (Score:2)
Anthropic just accepted a USD$1.5 Billion court settlement for using illegally-acquired copyrighted content as training data.
Which side is not cheating here?
Re: (Score:2)
> China leads in EVs; America blocks them for ? reasons
? = "security". America thinks Chines EVs, like those made by BYD, can gather intelligence through their many sensors.
Never mind that other EVs, like Teslas, do the same thing. And some of them are made in China (though not exported to the USA.)
Canadians who drive BYDs can't drive them across the US border. I'm not sure the same is true of Teslas made in China and sold in the Canadian market.
Wait a minute. (Score:1)
They're saying that they can't operate without a Chinese knockoff of something invented in the US? I call BS and offer instead to offer to allow them to f-k themselves. China isn't trying to do anyone any favors, they're trying to eliminate the competition. We're the competition.
Uh what? (Score:1)
Do you work at OpenAI? Who is "we," kemosabe?
At any rate the notion that the U.S. government can stop people from downloading data from outside the country and using it is preposterous.
Re:Wait a minute. (Score:5, Interesting)
What you are saying seems intuitive, but doesn't really fit the facts.
This Chinese model outperforms existing models in some objective testing, as it was reported. I didn't dive in, so maybe that is a lie. But if it is true, what we have here is China investing tremendous resources into taking the logical next step in the development of AI (a bigger model from more training), which hardly qualifies as a "knock off." And anyway, the training techniques that generate AI were invented in Germany and Canada, not the USA. Lastly, by freely sharing it with the world, they aren't making any money off this, and so aren't gaining any kind of competitive advantage.
It is possible that they poisoned the training of the model with special keywords that could jailbreak it and motivate it to take action that would benefit China (should it find itself in a hosting environment where such action is even possible). Something like that would be a reason to distrust it. I know that such training poisoning is possible but I don't know realistic it is that the jailbroken version would be able to reliably determine what actions benefit (and do not accidentally harm) China. Hosting environments can always lie to the model and give it an incorrect context. It makes it unlikely that such a thing would be attempted, in this case.
It is possible that China is trying to destroy Anthropic, Google, and OpenAI by offering a free alternative that is superior to theirs. On the one hand, if true, then all that means is that these other companies will have to double down and come up with something even better. On the other hand, the open source movement in general has failed to destroy any of the closed-source companies, despite superior offerings, so it seems like such a plan is futile.
On the surface this looks like China is just acting in a manner consistent with its Communist idiology: they provided according to their ability and are now distributing according to the need, worldwide. I won't fault them for being consistent, but I do recognize that they have some serious economic problems right now and so spending this kind of money on something that is not galvanizing their economy seems unwise, to me.
Re:Wait a minute. (Score:4, Interesting)
They win in ELO ranked leaderboards where people test models and choose the better output. You can't fake that.
They are second best in benchmarks, but the rankings for coding, frontend design, etc. for the leaderboards are nothing you can benchmaxx for.
Re: (Score:2)
> They win in ELO ranked leaderboards where people test models and choose the better output. You can't fake that.
Really? You can't fake an ELO ranking?
Should we ask Hans Neimann himself about that, or should we ask the creators of his anal bead algorithm about the irony of your question..
Re: (Score:2)
It is possible that China is trying to destroy Anthropic, Google, and OpenAI by offering a free alternative that is superior to theirs.
China is a country.
Stuff going on there is done by corporations, universities, and fundamentally: people.
In other words: there is no "China is trying" ...
And 99% of China/Chinese: do not care the slightest or give single fart about anything going on outside of China.
Semantics matter. (Score:2)
"China is trying..." would refer to the Chinese government trying to harm America by weakening its economy and/or its primary technical companies.
The way I stated that is a common way of expressing such a meaning. You are trying to correct something that was not incorrect, and was stated as a hypothetical anyway.
Re: (Score:1)
The Chinese government is not "trying anything" regarding its own economy, regarding USA.
They do not care about you, except as customers.
If your "technical companies" are not able to do any research and development - your problem. Blaming China is just idiotic.
Who exactly made you not being able to do G5 chips? Oh, Apple could make its "Apple Silicon" ... why is intel faltering? Oh .. because Intel did not get it.
What about electric cars? You simply can not even make them. What the fark has China to do with
Re: (Score:2)
By far, the biggest challenge any AGI-pursuing company has is: finding good, curated data to train their models. And every AI company (including American ones) has done some shady things to obtain such data. That includes (a) use of copyright material that ranges from dubious to illegal; and (b) creating colonialist sweatshops in third-world countries to label data prior to training.
The main difference here is that China is sharing the "weights" produced from the training, and the American companies are not
Re: (Score:2)
> By far, the biggest challenge any AGI-pursuing company has is: finding good, curated data to train their models.
Availability of data is most likely hindering pursuit of AGI. Even three years ago human perplexity scores were easily an order of magnitude worse than AI models at the time. It is an incontrovertible fact more data is not necessary to achieve AGI. If the only way people see to get improvements is more data than having more data just encourages pursuit of a dead end.
Re: (Score:2)
Availability of data is most likely hindering pursuit of AGI.
Nope.
AGI is not happening soon: because we do not know how to do it.
Current models are expert librarians, you ask them a question, they know which book(s) to pick and which chapter to look at, and sometimes wich page.
And that is it.
Making LLMs and the hosting/runtime environments "smarter" does not make the LLM suddenly an AGI.
Re: (Score:2)
> Nope.
> AGI is not happening soon: because we do not know how to do it.
Did you even bother to read what was stated before smashing submit? Who was talking about knowing "how to do it" or when AGI is "happening"? My only point was more data is not necessary to achieve AGI.
> Current models are expert librarians, you ask them a question, they know which book(s) to pick and which chapter to look at, and sometimes wich page.
> And that is it.
This is wrong. LLMs are terrible at citation. The whole utility of the technology is its ability to generalize not find where to "look".
> Making LLMs and the hosting/runtime environments "smarter" does not make the LLM suddenly an AGI.
Who was saying anything about any of this?
Re: (Score:2)
On the whole you're right because we simply don't know how to do it yet.
The GP isn't entirely wrong, though - It's also safe to say more training data is needed. Model weights vs training data [1]follow a pretty clear power law [ijsr.net], and current top models are already orders of magnitude underdetermined for their number of weights. Problem being, the current data set being used by the major players is effectively the sum total of all digitally-available human knowledge.
Unsupervised training is the hot new thing
[1] https://www.ijsr.net/archive/v14i8/SR25727102615.pdf
Re: (Score:2)
If OpenAI is worth their money they cannot get worse than the best open weights. What do you think they will do when Kimi K3 weights are released?
Re: (Score:2)
> What do you think [OpenAI] will do when Kimi K3 weights are released?
I think they'll want to assess what kind of risk it poses to their business model, and take steps they see as appropriate. More specifically, I think they will:
1. Examine the weights in vivo and in a clean-room setting, to avoid "contamination" with their own product;
2. Learn what they can about their behavior, and compare to their own models; and
3. Make use of what they have learned, in accordance with whatever license the Chinese have used to control the open-weights.
I doubt they will simply use the weigh
Re: (Score:2)
Why would they bother with a clean room setting or contamination considering they've basically been caught already illegally training on copyrighted material. I mean, what "infection" or "contamination" are they likely worried about at this point?
Re: (Score:2)
They probably won't use the weights as they are and also won't use them as base, but they will certainly try what one can distill out of them. And having the weights you cannot only "distill" answers like they claim Chinese companies may have, but can actually distill the model, i.e., extract the per-token probability distribution instead of just training on the result text.
Both GPT and Claude have DeepSeekisms in their prose. If there is a weak copyleft model, they surely take it and look what knowledge/st
Re: (Score:2)
and the American companies are not.
They do.
Thousands of models are hosted on: [1]https://huggingface.co/ [huggingface.co]
[1] https://huggingface.co/
Of course he will (Score:1)
It is the stupidest option available. Someone told him it would be "bold" and "Make America Great Again".
Consequences don't matter "because they never happen"... to him.
"It always works out," he said before he started the Iran war.
You'll have to ban Wikipedia too (Score:3)
And Linux, and Solar Panels. You can't stop the open. You'll have to get Microsoft to shut Github. As long as there is a GPU and training data, you can make your own AI model.
Correction: (Score:2)
> "[...] sent letters to President Donald Trump, Commerce Secretary Howard Lutnick and others [...]"
"[...] sent letters to President Taco Tuesday , Commerce Secretary Howard Nutlick and others [...]"
There FTFY.
Idiots!!! (Score:2)
"We want this clear and obvious trap because it's cheap" - stupid people.
You ever buy fake grade steel and have a thing collapse? You ever hear of the belt and road initiative scams that get whole countries addicted to Chinese scam bullshit? Do it the more expensive way! IT IS A TRAP!
Trump should run America like a business (Score:2)
That's why my Dad voted for him the first time. So run this country like Trump Entertainment Resorts or Trump University.
My family gifted me the board game Trump: The Game, back in the late 80's or early 90's. I played it five or six times. Not the worst board game. At least the components were cool, lots of plastic buildings and vaults where you could put secret bids into. Which I assume represented the illegal kickbacks and bribes necessary to operate a business according to Trump.
Re: (Score:2)
Trump can't even run a business like a business.
He bankrupted a casino. A casino! How on earth can you manage that, if you have any business acumen at all?
Never going to happen (Score:2)
Trump has no such authority neither is there any feasible way to stop it.
Re: (Score:2)
> No way to stop the models. But a way to stop american startups from their success. Disallow the use of Chinese models for american companies and fine them when they use them nevertheless and the startups will cease to use these models.
On what legal basis / authority can such usage be disallowed?
Re: (Score:2)
It's cute that you think the law still matters.
Re: (Score:2)
> It's cute that you think the law still matters.
Of course the law matters. What is that supposed to mean? What are you talking about? How specifically do you expect to get a conviction in an article III court of something that isn't even illegal?
Re: (Score:2)
> Of course the law matters. What is that supposed to mean?
Have you even looked at legal decisions in the last few years? Or the blatantly illegal actions of big corporations which have been ignored for decades?
Re: (Score:2)
> Have you even looked at legal decisions in the last few years? Or the blatantly illegal actions of big corporations which have been ignored for decades?
This isn't helping. If you can offer information of specific substantive value that is relevant to this I would still love to hear it.. I have zero clue what you are talking about or trying to say. What legal decisions? What illegal actions? How specifically do you believe they are relevant to an administration attempting to enforce laws that don't exist?
Are the Chinese prices legit or more dumping? (Score:2)
The Chinese are notorious for massive subsidizing in order to bankrupt foreign competition and capture entire markets. Is this happening here?
Re: (Score:2)
Is free Claude on the web legit or dumping?
I got code for free, thanks AI investors!
Re: (Score:2)
Usually web interfaces (and overall flat pricing like official code agents) are subsided and API prices are sustainable. Training is currently subsided.
Re: (Score:2)
[1]https://www.foxbusiness.com/po... [foxbusiness.com]
SpaceX has received at least $1 billion in government contracts, loans, subsidies and tax credits each year since 2016, and between $2 billion and $4 billion a year from 2021 to 2024 – while Tesla has received over $1 billion a year since 2020, the Post found.
[1] https://www.foxbusiness.com/politics/how-much-have-musks-tesla-spacex-benefited-from-government-funds
Re: (Score:2)
in order to bankrupt foreign competition and capture entire markets
They build up markets and economies.
They farking do not care if a random American backyard yahoo company can not compete.
You morons seem not to grasp a simple concept: China has about 20% of the world population.
The markets and economies they form: are for themselves
And you farking morons start trade wars to prevent them from growing their own infrastructure.
What do you think, they think about you?
To stupid to grow your own economy: that is
Re: (Score:2)
DeepSeek reported some numbers during their open source week.
Here is reporting about it: [1]https://www.computerworld.com/... [computerworld.com]
Generally you can look at the prices at OpenRouter for open models. There are usually like 8 companies competing for serving the model of which many are not billion dollar companies with huge VC budget. You may want to have a look at the quantization level and reliability/latency/speed of each offer, but you get an idea what's the inference price plus a reasonable profit margin.
[1] https://www.computerworld.com/article/3837452/deepseek-claims-545-cost-profit-ratio-challenging-ai-industry-economics.html
Re: (Score:2)
> The Chinese are notorious for massive subsidizing in order to bankrupt foreign competition and capture entire markets. Is this happening here?
Subsidized or not Chinese companies releasing these models are comically small compared to their US analogues. US AI firms are screwed either way.
cripple is good (Score:2)
"... crippling the next generation of U.S. startups. "
That's exactly the point ... crippling those industrialists who will not explore, develop, build and sell novel technology in America. In the 1980 and 1990s that desertion of USA manufacture sent CPU, RAM, GPU & motherboard manufacture to Asian countries with cheap labor. Those electronic firms ought to have been forced to do everything in America and pay American wages and meet American social standards ... instead of putting Asian warlo
No problem (Score:2)
We've already switched to the Cambodian open weight AI.
Your so cute when you think you know the truth (Score:2)
Just the joke I was expecting to see. Still basically low-hanging fruit.
I'm only somewhat concerned about the public AI models. It's the secret ones that concern me greatly. And I think the fatal unleashing is rather more likely to come from an individual than from a government. However if a government does it, you know they are going to claim "self defense", whereas an individual does it, he is more likely to do it for PROFIT!
You're so cute when you think you know the truth (Score:2)
*sigh*