Some of these were very good decisions imho, from someone who spent two months in Chineese rural area around ~2019.
- Tutoring platforms were a plague on Chineese youth that increased the weight of their already _very_ heavy load (tbh, i think and education reform might have been preferable, this is a stopgap, but at least it is something).
- Ant group was offering predatory consumption loans to rural China, which to me felt a lot like the "revolver credits" that plagued my country in the 80s and 90s and pushed to many to suicide (the surname cam from their english name, "revolving credit", and because my countryside had a lot of hunting rifle available to whomever). Considering how rural china is mistreaded by Chineese state and general government (and imho this is a real weakness in China politics), having this group by a huge fine for their practice and a general debt forgivness was great. Curtailing Ant's power is also good.
- Stopping consolidation is a great way to keep a market free.
- Crypto companies: mining diverted power from villages who couldn't compete on purchasing power to mining wharehouses in some state. The ban is great for the rural population at least. Also, if that can curtail the birth of Chineese cryptobros, great for the mental health of the country.
That is about investor confidence, not company performance. The companies are for the most part still making boatloads of money, just not as much as investors naively expected to get from taking over a market with 1.4B people.
And even if foreign investors are more cautious now, there is plenty of money trapped by capital controls, so that it doesn't look like new tech companies have trouble raising capital anyway.
Anth/OpenAI simply wanted the government to pull the ladder after them and ban models from China.
Seems it blew in their faces and probably the new frontier models will be available only to a select few. Many people predicted this, only a naive person would believe that access to something with these capabilities would be decided by some dude in California.
As entertaining as the sheer Schadenfreude of the situation is, this is terrible for foreign peasants like myself. It no longer makes any sense to pay for America's frontier AI models. I'd be funding the training of models I will never be able to use.
GLM 5.2 is competitive with Opus 4.6. If the best model I'll ever get is Opus 4.8, then the choice is clear. I'll miss Opus.
There is an assumption that everyone is making here - that China will not do the same. It is entirely possible, that China restricts their frontier models - as and when they are developed - to only Chinese citizens. And India follows along.
IMO AI is different from everything else. It is a weapon as potent as nuclear. It is only natural that it be treated as one.
> wouldn't that cause both Anthropic and OAI to fire non-citizens?
They would do what the thousands of other companies do with their tens of thousands of engineers handling ITAR/EAR-regulated software/hardware every day: compartmentalize their workforces, buildings, and access.
because the administration has been repeating the same patterns over pretty much its entire existence.
Dont worry though, the rest of the entire world gets access to better chinese models :-), once they get a taste for those the US has lost their little trade game and the future truly belongs to China.
Its almost like they are serving it up on a silver platter.
ofc they are not, they are just betting all in their models will be better, which is unlikely. (just look at the chinese law and all the names atop of advanced AI papers...)
So them banning Fable for only non-Americans is what we non-Americans should expect to be the norm going forward? Way to build even more resentment abroad.
I'm very pro-west, but at this point okay, I guess the rest of us have to side with China, not because we remotely like it, but because they don't try to be quite so antagonistic to us in everything they do.
Expect the US to sanction non-US-controlled models and put sanctions on individuals, companies and countries that use them? They already do this with other things like oil.
Can you cite any examples of a US citizen being sanctioned for importing foreign technology (not exporting)? Please don't cite anything OFAC-related, it does not apply here.
It’s looking very fragile from a legal point of view. Ownership of compute and software freedom will be next k the chopping block after control of networks that’s occurring at the moment.
This isn't going to save you unless you're ok being a criminal. There is nothing stopping the government from making open source versions of these models equally controlled.
And given how willy-nilly they are operating I see no reason they won't clamp down on open source. All it takes is someone with connections/political contributions wakes up one day and realizes that open source is a threat to their power or bottom line and it will be declared an imminent threat with no oversight or debate.
They'll just make it a crime to run the models unless they authorize you (classifying it as a munition, like they tried to do with encryption), and if your power bill is suspicious you'll find yourself in jail.
Any company providing the models will be deemed a threat to national security.
Citizens were and are free to use the technology (cryptography and every other export-controlled item); your "power bill is suspicious, go to jail" FUD doesn't really track with history.
> Any company providing the models will be deemed a threat to national security.
Any company providing specifically-controlled models to foreigners would hypothetically be prosecuted.
There's a famous poem called "First They Came" about how slippery this slope can be in a heated political climate.
I don't believe for a second this ends with "foreigners", this is about setting up infrastructure for controlling the technology. Foreigners are just the current excuse.
Note that TFA mentions they are supposedly hand-picking access to whoever they want, based on whatever criteria they want, already.
Ah, invoking Godwin. "First they came" in 1976 when ITAR was first passed, or maybe "first they came" in the 1940s when we didn't export Proximity Fuzes, right?
Countries are free to prevent exports of technology. Equating export controls with the Holocaust is disgusting.
I did not bring in Godwin, but I guess he's here now :D.
I'm more trying to invoke GRRM. This is a Game of Thrones: biollionaire CEO's complain about each other to the government to get their competitors blocked/tripped up with acts of fiat, which is what happened with Anthropic.
And in the linked post, it says GPT-5.6 access decisions are supposedly just hand picked.
That’s not true at all. While not as good as proprietary models they are still very good and can do A LOT, certainly more than their cost would make it seem.
It’s only a matter of time before companies start to acknowledge the huge cost of tokens and look for a cheaper alternative with basic cost-benefit analysis.
My F500 company is getting local infrastructure going to host open models and I’m sure many will just switch to bedrock + the best open models.
It’s foolish for companies to let three companies dictate the price of tokens, I just don’t think they are aware of this now by and large.
Without access to leading models, I think open source LLM development will also slow down. I'm not sure which portion of their success right now is due to RLAF and distillation but it's certainly not zero.
This will be the end of the US's short-lived AI supremacy. OpenAI and Anthropic are already wildly unprofitable, cutting off the world-wide income stream is just fucking bad business.
Don't worry, their pals in the government will bail them out.
But it is odd that this administration has learned absolutely nothing about the mid- to long-term effects of export restrictions on other countries' ability to compete with the US.
Mid and long term effects will come with next administration - which can be blamed for the failure (even if it has nothing to do with it) -> so those who caused the problem can be voted back to power.
In all seriousness, I can’t believe the AI firms are abiding by this peacefully. If I truly loved my company, and I felt we were on the bleeding edge of incredible, life changing products, why would I allow my company to be set up for failure by remaining somewhere that clearly wants control over the sovereignty.
The US gov sees these AI companies as bartering power, not as innovation. Wouldn’t you as a parent always want what’s best for your child, not for yourself?
It also feels like they can’t just relocate out of the country, as the administration will surely sanction anyone from business within the country again. These firms are so over inflated with evaluations and opex, they’ve dug themselves into a corner.
This is not to say regulation does not exist in any other country, but it’s clear now after what’s happening at Anthropic + OAI that the US gov has taken these companies hostage.
This is only further playing into the hands of open source and the outside models; the US gov is going to be to blame for when they all lose the race to low cost/free.
> In all seriousness, I can’t believe the AI firms are abiding by this peacefully. If I truly loved my company, and I felt we were on the bleeding edge of incredible, life changing products, why would I allow my company to be set up for failure by remaining somewhere that clearly wants control over my sovereignty.
So, locate in China, where every company of importance is essentially required in practice to maintain ties to the CCP?
I personally think the US has gone too far with its use of export controls and sanctions as a political tool, but it's foolish to believe that it's different anywhere else on the planet.
In China, it has even been reported that top AI talent is restricted from overseas travel.
Bottom line: if you're working on cutting-edge technology that is deemed to be of critical national security importance and has military or dual use implications, you're going to be a hostage no matter where you go.
You wrongly assumed I implied these firms relocate to China. We are all aware of how China operates and controls its assets.
AI has long existed in many countries around the world without this type of behavior from the government. Deepmind in the UK, Mistral in France, DeepL in Germany - the governments don’t seem to be forcing employees to get their deploys approved by a government official.
My argument is that the US gov does not like that these companies have too much influence which they do not feel they can mandate. It’s slowing the entire country down at a very critical sink or swim inflection point in this tech.
I think it’s pretty clear why they’re abiding by this:
-the US is the only place where you can raise the kinds of money you need to run a lab like this.
-a government that won’t let you sell products to customers abroad will probably object even more to you moving abroad.
Even if you made the move abroad, that government might no longer let you access US data centers.
-This basically affects OpenAI and Anthropic, which make the only LLMs most people consider frontier nowadays. Since most open weights models rely on distillation of frontier models, it may genuinely entrench those companies more.
It may be playing into the hands of open source OAI/Anthropic dependencies start to look more dangerous, but it also makes building better OSS models harder.
The advantages the AI labs rely on might be less durable than a proprietary process in industrial manufacturing, but it’s still meaningful.
I think the bigger reckoning will come from a different angle: tokens will eventually need to cover cost.
That will likely mean multiplying prices compared to today. And companies already complain now!
Model orchestration and smaller models that can run locally or cheaply will become more important in my opinion.
Right now, you can still default to GPT/Claude and it’s kind of fine, but that will have to change.
The elephant in the room is that the US AI firms should not be as valuable as they are. They should not require the sort of capital they are seeking, the amount of employees, the amount of offices and resources..but they are so steeped in investor interests - why stop being fed?
Many Americans want AI to fail. The US gov wants to control AI. The AI companies are running out of things to do, and are shipping product after product after product to keep the perceived productivity narrative alive.
At this rate I would not be surprised to see an OAI/Anthropic merger just to throw everything AI the US has to offer to the global markets.
Whether they’re over-valued and over-resourced is a big question. I think that will be answered when eventual price hikes happen and people shift which AI they use and/or what they use it for.
We’re still in the “5$ airport Ubers thanks to VC money” era of AI
This will be exactly as effective as the BBC's efforts to ensure only UK taxpayers are allowed to stream Doctor Who from BBC servers on Christmas morning.
GLM-5.2 is currently the best open-weight model for development. It's not as good as the current American SOTA models, but if you wrote code with US SOTA models four months ago, you can write code with GLM-5.2 today.
DeepSeek 4 is a good model for many tasks, but I think it currently lacks the post-training required to become a genuinely great coding model.
Wow.. Okay so it's official now that the playbook is "we will try to prevent anyone who we don't like to use advanced tech".
I understand if its military hardware and software, that's the property of the US government however this is the property of a private company.. Now seemingly being commandeered and issued at the will of the government, sounds very Russian/Chinese to me.
Is there a precedent for this before in a democratic country ?
The overwhelming majority of export-controlled items are made by private corporations: the US government itself makes exceedingly little in comparison.
The missiles Raytheon makes are export-controlled too, and they're not somehow "property of the US government" - this isn't China.
Is this just upsetting because it's a product you want to enjoy?
> Is there a precedent for this before in a democratic country ?
Try every weapons system, encrypted radio system, FPGAs with high-bandwidth transceivers, lithography equipment, etc. etc. etc. There's plenty of precedent.
> Is this just upsetting because it's a product you want to enjoy?
No, infact I'm a proponent of open models and being able to run them locally, it just feels strange that a consumer product would be under the same restrictions as military grade equipment and tech which is specifically designed for warfare.
> Try every weapons system, encrypted radio system, FPGAs with high-bandwidth transceivers, lithography equipment, etc. etc. etc. There's plenty of precedent.
If it's the same equivalent then my issue is just that, it feels like trying to restrict the useage of RSA because it could be used by bad actors.
> If it's the same equivalent then my issue is just that, it feels like trying to restrict the useage of RSA because it could be used by bad actors.
RSA was practically impossible to control (an implementation is what, 100 lines in any language?) and the global benefits outweighed the cost and futility associated with restrictions.
AI laboratories with hundreds of billions of dollars in funding aren't cropping up in every country in the world, and their products and services are easily controlled and not easily replicated.
"Freedom of commerce" doesn't mean "unchecked globalism" - there are plenty of dual-use items that only friendly countries or citizens can obtain (and within those categories, there aren't any further restrictions besides "don't share.")
The UK is a lot more compassionate about people’s wishes, it’s not nearly as bureaucratic and polarizing “democracy” as the US. Laws in the UK are passed quickly, and feedback is always considered. Whether you agree or not on the regulation is another discussion.
Probably you're right overall but that doesn't apply to anyone who chooses to want to educate their kids in a non-taxpayer funded State school. Around 100–105 independent schools were reported as having ceased operations after the UK government introduced 20% VAT on private school fees from January 2025. Some may feel (I would not dare suggest it) that the current government is on a mission to close them all up unless they attract sufficiently rich parents like Eton. Closing the latter would be news indeed. However - exit Exeter Cathedral School after 847 years, which taught Charles II's composer and Coldplay's Chris Martin. It's closing with financial difficulties which have beset the sector in general since charges were introduced.
If it's the case then software engineers still have the same place as pre-ClaudeCode era, because 4.8 and 5.5 are damn good at algo but notoriously bad at architecture and coordination.
Thousands of American engineers all over the country (most of whom probably aren't on Hacker News) work with ITAR/EAR-regulated software and hardware every single day: these regulations are really not difficult to abide if you're a citizen.
Except it isn't zero-sum thinking: the rest of the world can have the scraps, and as long as the scraps are marginally better than the rest of the world's offerings, they will sell.
> What are they going to do? Start their own Anthropic? Go for it. Why is every other country in the world entitled to American technology by default?
Because American tech companies make a lot of money from outside of the US. For instance, 1/4 of all Apple revenues are from Europe, and 1/5 from China and China-claimed territories. Only around 40% are from the Americas (so not even the US exclusively).
Would American tech companies be as successfull without ~half their revenues?
In any case, it doesn't matter, the cat is out of the bag. Nobody sane and non-American would trust American frontier labs, because their models can be yanked at will by whoever is in the White House. It would be suicidal to rely on them for critical business or developer workflows. So your options are to go with Mistral or open source Chinese models, hosted within your environment, with the added benefits of being able to control the costs and being able to fine tune the models to better work for you.
I'm referring to OpenAI and Antropic - would they be successfull with ~40-50% of their potential market?
And iPhones, not really. But you can bet your ass that every business purchasing software in Europe is at least considering the geopolitical risks of buying American, and thinking of alternatives. Doesn't mean they'll all stop buying American software any time soon, but the shift has already started.
> I'm referring to OpenAI and Antropic - would they be successfull with ~40-50% of their potential market?
You presume that every single product they sell will be restricted: this is unrealistic. The rest of the world can have the gimped models, and as so long as they're better than other offerings, the revenue will flow - which is exactly what happens with countless other dual-use goods.
One more wake up call for anyone outside USA, especially Europe. AI will be weaponized, on the battle ground too, but the bigger battle will be fought in the industry competition. Those who have access to state of the art models will have advantage over those who does not.
Hopefully open-weight models will catch up, hopefully we, as the people, engineers will find the way to maintain those open-weight models on pair with the closed ones.
I try to be optimistic, as we won some battles, against all odds, Linux is flourishing, open source solutions are mainstream.
A consortium will train a 400B-class model and get 2.5% of time of the EuroHPC infrastructure (~2000 PFLOPS datacenters). So, even if the Chinese take away the open source there will be some models. Probably not Mythos quality yet though.
Europe is in the worst spot right now, because even if open source is the future, there is not enough European-owned datacenters even for inference. Not to mention that China could pull the rug on these models at any moment just like the US did.
Keep your **** models to yourselves.... the world really has moved on to open models which can give you good enough results at a fraction of the cost and zero BS licensing.
Don't get me wrong: I'm all for open models, but I think it will get more and more difficult to distil-train them without (legitimate) access to frontier models.
I’m not sure, because the same thing happened with facebook advertising restrictions during the 2018 elections and nowadays there’s a whole black market for fake ad accounts.
If anything I bet these people will just use their knowledge to make even more money reselling tokens.
Personally, I find it rather humorous that we've moved from the fear that AI generated output would corrupt training to the idea that it is essential to training. Reality itself has not just a left bias but a bias to fundamentals. Bootstrap from fundamentals without introducing arbitrary error and you have the superior system; it just may not be highly compatible with a trash ecosystem.
I mean, I'm not sure that's the correct read on this.
If you want an Opus class model, it makes sense that you would train on what Opus outputs. But, if you want something better than Opus, training on the same data that Opus was trained on with the same architecture will only result in an Opus class model. Then, if your dataset also contains Opus outputs, many of which are wrong, then it makes sense that the model would have reduced performance.
All this to say that I don't think there's such a thing as a "Model Collapse," but there likely is a "Model Stagnation."
Remember how China turned its tech industry into a smoking ruin - make making them all submit to political priorities:
Ant Group: China halted Ant’s IPO and forced a restructuring
Alibaba: China fined and politically disciplined Alibaba
Didi: China punished Didi after its US listing by removing its apps, freezing users, forcing delisting
Tutoring platforms: banned profit from core school-subject tutoring.
Tencent gaming: restricted youth gaming froze approvals
NetEase and gaming companies: licence freeze stopped game companies from shipping games.
Meituan: fined Meituan and forced changes to its labour and platform model.
Huya/DouYu: blocked Tencent’s game-streaming merger, stopping commercial consolidation in a major entertainment market.
Boss Zhipin / Full Truck Alliance: froze new users after listging in the US
Crypto companbies: banned crypto trading and mining, forcing exchanges offshore.
Think it's not happening to the US?
tourism - people afraid to visit
tariffs - wrecking ball to all businesses
defence - why would anyone buy US weapons after Greenland and Canada
internet clouds - Greenland made Europe decide that the US clouds can't be trusted, now sovereign computing matters and MS/AWS/Google are feeling it
finance - no one trusts the US not to turn people into "non members of global society" by banning them from visa and credit card and banking systems
From my perspective curtailing Ant's plans was positive regulatory action.
Political priorities and good governance is why we have government.
Some of these were very good decisions imho, from someone who spent two months in Chineese rural area around ~2019.
- Tutoring platforms were a plague on Chineese youth that increased the weight of their already _very_ heavy load (tbh, i think and education reform might have been preferable, this is a stopgap, but at least it is something).
- Ant group was offering predatory consumption loans to rural China, which to me felt a lot like the "revolver credits" that plagued my country in the 80s and 90s and pushed to many to suicide (the surname cam from their english name, "revolving credit", and because my countryside had a lot of hunting rifle available to whomever). Considering how rural china is mistreaded by Chineese state and general government (and imho this is a real weakness in China politics), having this group by a huge fine for their practice and a general debt forgivness was great. Curtailing Ant's power is also good.
- Stopping consolidation is a great way to keep a market free.
- Crypto companies: mining diverted power from villages who couldn't compete on purchasing power to mining wharehouses in some state. The ban is great for the rural population at least. Also, if that can curtail the birth of Chineese cryptobros, great for the mental health of the country.
China tech industry is smoking ruin? On what planet are you living?
I suppose we should bookmark this for the next time that HN claims that the rise of China's tech industry is inevitable.
Tech crackdowns rid China of entrepreneurial capitalism https://eastasiaforum.org/2023/09/12/tech-crackdowns-rid-chi...
Why China crushed its tech giants https://www.wired.com/story/china-tech-giants-policy/
Why Big Tech May Never Recover in China https://time.com/6973119/china-big-tech-crackdown-backfiring...
Beijing can’t afford another crackdown on its tech companies https://www.cnbc.com/2026/03/11/china-cant-afford-another-cr...
That is about investor confidence, not company performance. The companies are for the most part still making boatloads of money, just not as much as investors naively expected to get from taking over a market with 1.4B people.
And even if foreign investors are more cautious now, there is plenty of money trapped by capital controls, so that it doesn't look like new tech companies have trouble raising capital anyway.
US citizens to remain nonviolent at any cost, issue strongly worded internet comments, and find themselves a little less free every day.
While laughing at the stereotype of French being on the street all the time
I do respect the French. They've proven, time and time again, that if you fuck with the people, heads will roll...
The onion finds itself in a peculiar spot today
"government needs to step in and regulate ai"
"wait, not like that"
Anth/OpenAI simply wanted the government to pull the ladder after them and ban models from China.
Seems it blew in their faces and probably the new frontier models will be available only to a select few. Many people predicted this, only a naive person would believe that access to something with these capabilities would be decided by some dude in California.
As entertaining as the sheer Schadenfreude of the situation is, this is terrible for foreign peasants like myself. It no longer makes any sense to pay for America's frontier AI models. I'd be funding the training of models I will never be able to use.
GLM 5.2 is competitive with Opus 4.6. If the best model I'll ever get is Opus 4.8, then the choice is clear. I'll miss Opus.
There is an assumption that everyone is making here - that China will not do the same. It is entirely possible, that China restricts their frontier models - as and when they are developed - to only Chinese citizens. And India follows along.
IMO AI is different from everything else. It is a weapon as potent as nuclear. It is only natural that it be treated as one.
im crying bro snuck in india
Why do I get the feeling the administration is doing this to buy a position in the AI companies before they go public.
If non US citizens shouldn't have the models - wouldn't that cause both Anthropic and OAI to fire non-citizens?
> wouldn't that cause both Anthropic and OAI to fire non-citizens?
They would do what the thousands of other companies do with their tens of thousands of engineers handling ITAR/EAR-regulated software/hardware every day: compartmentalize their workforces, buildings, and access.
because the administration has been repeating the same patterns over pretty much its entire existence.
Dont worry though, the rest of the entire world gets access to better chinese models :-), once they get a taste for those the US has lost their little trade game and the future truly belongs to China.
Its almost like they are serving it up on a silver platter.
ofc they are not, they are just betting all in their models will be better, which is unlikely. (just look at the chinese law and all the names atop of advanced AI papers...)
So them banning Fable for only non-Americans is what we non-Americans should expect to be the norm going forward? Way to build even more resentment abroad.
I'm very pro-west, but at this point okay, I guess the rest of us have to side with China, not because we remotely like it, but because they don't try to be quite so antagonistic to us in everything they do.
>because they don't try to be quite so antagonistic to us in everything they do.
Just because the many headed dragon is trying to bite your sailors' heads doesn't mean you should pilot your ship into the whirlpool
I couldn't have put it better myself.
Expect the US to sanction non-US-controlled models and put sanctions on individuals, companies and countries that use them? They already do this with other things like oil.
Can you cite any examples of a US citizen being sanctioned for importing foreign technology (not exporting)? Please don't cite anything OFAC-related, it does not apply here.
Oil isn't made out of information and cannot be transmitted via a speech act.
Open source is looking great right now
It’s looking very fragile from a legal point of view. Ownership of compute and software freedom will be next k the chopping block after control of networks that’s occurring at the moment.
Less so in EU than in US.
I would not be so confident, though I certainly hope!
This isn't going to save you unless you're ok being a criminal. There is nothing stopping the government from making open source versions of these models equally controlled.
And given how willy-nilly they are operating I see no reason they won't clamp down on open source. All it takes is someone with connections/political contributions wakes up one day and realizes that open source is a threat to their power or bottom line and it will be declared an imminent threat with no oversight or debate.
It's looking good until you start to see the US gov forcing cloudflare to block hugging face and others.
They'll just make it a crime to run the models unless they authorize you (classifying it as a munition, like they tried to do with encryption), and if your power bill is suspicious you'll find yourself in jail.
Any company providing the models will be deemed a threat to national security.
No need to block the download.
Citizens were and are free to use the technology (cryptography and every other export-controlled item); your "power bill is suspicious, go to jail" FUD doesn't really track with history.
> Any company providing the models will be deemed a threat to national security.
Any company providing specifically-controlled models to foreigners would hypothetically be prosecuted.
There's a famous poem called "First They Came" about how slippery this slope can be in a heated political climate.
I don't believe for a second this ends with "foreigners", this is about setting up infrastructure for controlling the technology. Foreigners are just the current excuse.
Note that TFA mentions they are supposedly hand-picking access to whoever they want, based on whatever criteria they want, already.
Ah, invoking Godwin. "First they came" in 1976 when ITAR was first passed, or maybe "first they came" in the 1940s when we didn't export Proximity Fuzes, right?
Countries are free to prevent exports of technology. Equating export controls with the Holocaust is disgusting.
I did not bring in Godwin, but I guess he's here now :D.
I'm more trying to invoke GRRM. This is a Game of Thrones: biollionaire CEO's complain about each other to the government to get their competitors blocked/tripped up with acts of fiat, which is what happened with Anthropic.
And in the linked post, it says GPT-5.6 access decisions are supposedly just hand picked.
Why do they need to "force Cloudflare" to do anything?
Why wouldn't they just tell Hugging Face that they need to abide export restrictions directly - they're an American company?
Doesn't sound dystopian enough without a second compelled entity?
seriously, ordered more hardware this week, as it gets more dystopian every week
wondering when more people will raise their voice and get engaged
History shows that people generally start speaking out about things after it's too late to do so.
It's the year of the open source AI model is the new 'It's the year of the Linux Desktop'. It's not and never will be for 90% of people
That’s not true at all. While not as good as proprietary models they are still very good and can do A LOT, certainly more than their cost would make it seem.
It’s only a matter of time before companies start to acknowledge the huge cost of tokens and look for a cheaper alternative with basic cost-benefit analysis.
My F500 company is getting local infrastructure going to host open models and I’m sure many will just switch to bedrock + the best open models.
It’s foolish for companies to let three companies dictate the price of tokens, I just don’t think they are aware of this now by and large.
Well if us gov would block people from using windows or macos, then it may well be.
GLM 5.2 is competitive with Opus 4.6.
[dead]
Without access to leading models, I think open source LLM development will also slow down. I'm not sure which portion of their success right now is due to RLAF and distillation but it's certainly not zero.
Source: https://www.reuters.com/business/trump-administration-asks-o...
In a scenario where some breakthrough in fusion energy will be discovered I envision:
- instant, total world war if it's not coming from USA
- let's finish all oil's reserve first otherwise
This will be the end of the US's short-lived AI supremacy. OpenAI and Anthropic are already wildly unprofitable, cutting off the world-wide income stream is just fucking bad business.
Don't worry, their pals in the government will bail them out.
But it is odd that this administration has learned absolutely nothing about the mid- to long-term effects of export restrictions on other countries' ability to compete with the US.
You mean pension funds will bail them out after they IPO? :)
Mid and long term effects will come with next administration - which can be blamed for the failure (even if it has nothing to do with it) -> so those who caused the problem can be voted back to power.
Thanks to the US government for helping kill Anthropic and OpenAI by preventing them from recouping any R&D money from new models. Doing god's work.
+1 point to China!
In all seriousness, I can’t believe the AI firms are abiding by this peacefully. If I truly loved my company, and I felt we were on the bleeding edge of incredible, life changing products, why would I allow my company to be set up for failure by remaining somewhere that clearly wants control over the sovereignty.
The US gov sees these AI companies as bartering power, not as innovation. Wouldn’t you as a parent always want what’s best for your child, not for yourself?
It also feels like they can’t just relocate out of the country, as the administration will surely sanction anyone from business within the country again. These firms are so over inflated with evaluations and opex, they’ve dug themselves into a corner.
This is not to say regulation does not exist in any other country, but it’s clear now after what’s happening at Anthropic + OAI that the US gov has taken these companies hostage.
This is only further playing into the hands of open source and the outside models; the US gov is going to be to blame for when they all lose the race to low cost/free.
> +1 point to China!
Which, like the US, uses export controls when it finds them advantageous: https://nam.org/china-imposes-export-controls-on-u-s-mineral...
> In all seriousness, I can’t believe the AI firms are abiding by this peacefully. If I truly loved my company, and I felt we were on the bleeding edge of incredible, life changing products, why would I allow my company to be set up for failure by remaining somewhere that clearly wants control over my sovereignty.
So, locate in China, where every company of importance is essentially required in practice to maintain ties to the CCP?
I personally think the US has gone too far with its use of export controls and sanctions as a political tool, but it's foolish to believe that it's different anywhere else on the planet.
In China, it has even been reported that top AI talent is restricted from overseas travel.
https://www.thinkchina.sg/technology/china-tightens-control-...
Bottom line: if you're working on cutting-edge technology that is deemed to be of critical national security importance and has military or dual use implications, you're going to be a hostage no matter where you go.
You wrongly assumed I implied these firms relocate to China. We are all aware of how China operates and controls its assets.
AI has long existed in many countries around the world without this type of behavior from the government. Deepmind in the UK, Mistral in France, DeepL in Germany - the governments don’t seem to be forcing employees to get their deploys approved by a government official.
My argument is that the US gov does not like that these companies have too much influence which they do not feel they can mandate. It’s slowing the entire country down at a very critical sink or swim inflection point in this tech.
I think it’s pretty clear why they’re abiding by this:
-the US is the only place where you can raise the kinds of money you need to run a lab like this.
-a government that won’t let you sell products to customers abroad will probably object even more to you moving abroad.
Even if you made the move abroad, that government might no longer let you access US data centers.
-This basically affects OpenAI and Anthropic, which make the only LLMs most people consider frontier nowadays. Since most open weights models rely on distillation of frontier models, it may genuinely entrench those companies more.
It may be playing into the hands of open source OAI/Anthropic dependencies start to look more dangerous, but it also makes building better OSS models harder.
The advantages the AI labs rely on might be less durable than a proprietary process in industrial manufacturing, but it’s still meaningful.
I think the bigger reckoning will come from a different angle: tokens will eventually need to cover cost.
That will likely mean multiplying prices compared to today. And companies already complain now!
Model orchestration and smaller models that can run locally or cheaply will become more important in my opinion.
Right now, you can still default to GPT/Claude and it’s kind of fine, but that will have to change.
The elephant in the room is that the US AI firms should not be as valuable as they are. They should not require the sort of capital they are seeking, the amount of employees, the amount of offices and resources..but they are so steeped in investor interests - why stop being fed?
Many Americans want AI to fail. The US gov wants to control AI. The AI companies are running out of things to do, and are shipping product after product after product to keep the perceived productivity narrative alive.
At this rate I would not be surprised to see an OAI/Anthropic merger just to throw everything AI the US has to offer to the global markets.
Whether they’re over-valued and over-resourced is a big question. I think that will be answered when eventual price hikes happen and people shift which AI they use and/or what they use it for.
We’re still in the “5$ airport Ubers thanks to VC money” era of AI
> I can’t believe the AI firms are abiding by this peacefully.
They literally asked for this.
> AI firms are abiding by this peacefully
What are they going to do about it? Might makes right.
They've already done what little they can: pull access to their models wholesale rather than adopt an export compliance regime.
This will be exactly as effective as the BBC's efforts to ensure only UK taxpayers are allowed to stream Doctor Who from BBC servers on Christmas morning.
So, that DeepSeek thing, you are saying it's not that bad?
GLM-5.2 is currently the best open-weight model for development. It's not as good as the current American SOTA models, but if you wrote code with US SOTA models four months ago, you can write code with GLM-5.2 today.
DeepSeek 4 is a good model for many tasks, but I think it currently lacks the post-training required to become a genuinely great coding model.
Wow.. Okay so it's official now that the playbook is "we will try to prevent anyone who we don't like to use advanced tech".
I understand if its military hardware and software, that's the property of the US government however this is the property of a private company.. Now seemingly being commandeered and issued at the will of the government, sounds very Russian/Chinese to me.
Is there a precedent for this before in a democratic country ?
The overwhelming majority of export-controlled items are made by private corporations: the US government itself makes exceedingly little in comparison.
The missiles Raytheon makes are export-controlled too, and they're not somehow "property of the US government" - this isn't China.
Is this just upsetting because it's a product you want to enjoy?
> Is there a precedent for this before in a democratic country ?
Try every weapons system, encrypted radio system, FPGAs with high-bandwidth transceivers, lithography equipment, etc. etc. etc. There's plenty of precedent.
> Is this just upsetting because it's a product you want to enjoy?
No, infact I'm a proponent of open models and being able to run them locally, it just feels strange that a consumer product would be under the same restrictions as military grade equipment and tech which is specifically designed for warfare.
> Try every weapons system, encrypted radio system, FPGAs with high-bandwidth transceivers, lithography equipment, etc. etc. etc. There's plenty of precedent.
If it's the same equivalent then my issue is just that, it feels like trying to restrict the useage of RSA because it could be used by bad actors.
> If it's the same equivalent then my issue is just that, it feels like trying to restrict the useage of RSA because it could be used by bad actors.
RSA was practically impossible to control (an implementation is what, 100 lines in any language?) and the global benefits outweighed the cost and futility associated with restrictions.
AI laboratories with hundreds of billions of dollars in funding aren't cropping up in every country in the world, and their products and services are easily controlled and not easily replicated.
That was always the playbook
> Is there a precedent for this before in a democratic country ?
I'd argue US is not very democratic country given how many of what govt does goes against people's wishes. Same as UK
> I'd argue US is not very democratic country given how many of what govt does goes against people's wishes. Same as UK
That could be argued but the core principle is freedom of commerce and private companies get a lot of runway. This seems completely counter to tha.
"Freedom of commerce" doesn't mean "unchecked globalism" - there are plenty of dual-use items that only friendly countries or citizens can obtain (and within those categories, there aren't any further restrictions besides "don't share.")
The UK is a lot more compassionate about people’s wishes, it’s not nearly as bureaucratic and polarizing “democracy” as the US. Laws in the UK are passed quickly, and feedback is always considered. Whether you agree or not on the regulation is another discussion.
> Laws in the UK are passed quickly
Is that a feature or a bug?
Depends on how pessimistic you are I suppose.
Probably you're right overall but that doesn't apply to anyone who chooses to want to educate their kids in a non-taxpayer funded State school. Around 100–105 independent schools were reported as having ceased operations after the UK government introduced 20% VAT on private school fees from January 2025. Some may feel (I would not dare suggest it) that the current government is on a mission to close them all up unless they attract sufficiently rich parents like Eton. Closing the latter would be news indeed. However - exit Exeter Cathedral School after 847 years, which taught Charles II's composer and Coldplay's Chris Martin. It's closing with financial difficulties which have beset the sector in general since charges were introduced.
Not really. The UK is run (mostly) by career politicians, they really do not care.
This is for the preview period, but it's not a good sign. Opus 4.8 may be the last frontier model available to the masses...
From US companies that is.
If it's the case then software engineers still have the same place as pre-ClaudeCode era, because 4.8 and 5.5 are damn good at algo but notoriously bad at architecture and coordination.
Yes, we will get a crippled version of Mythos, 5.6 and future models, while the chosen few will have unfettered access.
Thousands of American engineers all over the country (most of whom probably aren't on Hacker News) work with ITAR/EAR-regulated software and hardware every single day: these regulations are really not difficult to abide if you're a citizen.
And what about the rest of the world? I can't imagine US partners will abide this for long.
They get the dual-use scraps or whatever China is hawking.
Being told "no" is never fun, but the regulations are not hard to comply with (despite what Anthropic might have you believe.)
> I can't imagine US partners will abide this for long.
What are they going to do? Start their own Anthropic? Go for it. Why is every other country in the world entitled to American technology by default?
> Why is every other country in the world entitled to American technology by default?
This kind of zero-sum thinking is what is killing the US's global influence right now.
Except it isn't zero-sum thinking: the rest of the world can have the scraps, and as long as the scraps are marginally better than the rest of the world's offerings, they will sell.
You do see how that is not going to work, right?
Seems to work for every other export-controlled good and service.
Oh, ok, I get it. Sorry, I'm bad at detecting sarcasm on the Internet.
> What are they going to do? Start their own Anthropic? Go for it. Why is every other country in the world entitled to American technology by default?
Because American tech companies make a lot of money from outside of the US. For instance, 1/4 of all Apple revenues are from Europe, and 1/5 from China and China-claimed territories. Only around 40% are from the Americas (so not even the US exclusively).
Would American tech companies be as successfull without ~half their revenues?
In any case, it doesn't matter, the cat is out of the bag. Nobody sane and non-American would trust American frontier labs, because their models can be yanked at will by whoever is in the White House. It would be suicidal to rely on them for critical business or developer workflows. So your options are to go with Mistral or open source Chinese models, hosted within your environment, with the added benefits of being able to control the costs and being able to fine tune the models to better work for you.
> Would American tech companies be as successfull without ~half their revenues?
Good luck with "if you don't let us use your AI technology, we wont allow iPhones in" - go for it.
Yep, we all can play tit for tat.
Needlessly patriotic and confrontational.
I'm referring to OpenAI and Antropic - would they be successfull with ~40-50% of their potential market?
And iPhones, not really. But you can bet your ass that every business purchasing software in Europe is at least considering the geopolitical risks of buying American, and thinking of alternatives. Doesn't mean they'll all stop buying American software any time soon, but the shift has already started.
> I'm referring to OpenAI and Antropic - would they be successfull with ~40-50% of their potential market?
You presume that every single product they sell will be restricted: this is unrealistic. The rest of the world can have the gimped models, and as so long as they're better than other offerings, the revenue will flow - which is exactly what happens with countless other dual-use goods.
One more wake up call for anyone outside USA, especially Europe. AI will be weaponized, on the battle ground too, but the bigger battle will be fought in the industry competition. Those who have access to state of the art models will have advantage over those who does not.
Hopefully open-weight models will catch up, hopefully we, as the people, engineers will find the way to maintain those open-weight models on pair with the closed ones.
I try to be optimistic, as we won some battles, against all odds, Linux is flourishing, open source solutions are mainstream.
There are some steps in the good direction:
https://digital-strategy.ec.europa.eu/en/news/commission-sel...
A consortium will train a 400B-class model and get 2.5% of time of the EuroHPC infrastructure (~2000 PFLOPS datacenters). So, even if the Chinese take away the open source there will be some models. Probably not Mythos quality yet though.
Europe is in the worst spot right now, because even if open source is the future, there is not enough European-owned datacenters even for inference. Not to mention that China could pull the rug on these models at any moment just like the US did.
[dead]
Keep your **** models to yourselves.... the world really has moved on to open models which can give you good enough results at a fraction of the cost and zero BS licensing.
> the world really has moved on to open models
Don't get me wrong: I'm all for open models, but I think it will get more and more difficult to distil-train them without (legitimate) access to frontier models.
I’m not sure, because the same thing happened with facebook advertising restrictions during the 2018 elections and nowadays there’s a whole black market for fake ad accounts.
If anything I bet these people will just use their knowledge to make even more money reselling tokens.
As if all progress done in open models is because of distilling...
People have no idea and everybody pretends to be an expert and ignore how good China is on AI research
Personally, I find it rather humorous that we've moved from the fear that AI generated output would corrupt training to the idea that it is essential to training. Reality itself has not just a left bias but a bias to fundamentals. Bootstrap from fundamentals without introducing arbitrary error and you have the superior system; it just may not be highly compatible with a trash ecosystem.
I mean, I'm not sure that's the correct read on this.
If you want an Opus class model, it makes sense that you would train on what Opus outputs. But, if you want something better than Opus, training on the same data that Opus was trained on with the same architecture will only result in an Opus class model. Then, if your dataset also contains Opus outputs, many of which are wrong, then it makes sense that the model would have reduced performance.
All this to say that I don't think there's such a thing as a "Model Collapse," but there likely is a "Model Stagnation."