I hope this doesn't become the new norm where government becomes the bottleneck for innovation in the AI space.
It's worrying that with no formal and transparent policy framework that the government will be picking winners and losers and stifling innovation.
There's been no public policy, executive order, legislation, or otherwise on this, I wonder if anyone has filed FOIA requests for these decisions or the conversations between the Executive Branch and AI companies.
Indeed, I find quite ironic that some people in tech in the US complain about EU "regulations first" approach, but then their government seem to arbitrarily stop things from being released because, well, there is no established policy on safety guarantees or other similar aspects.
I see it too, but worth noting that this is basically unprecedented at least within the last 25 years; I think you have to go back to export controlled cryptography for another example of this kind of abrupt and targeted regulation.
Huawei, Foreign gambling sites were banned on dubious reasons in 2006 (in reality American companies weren't as competitive and las Vegas needed to be protected), Japanese electronic tariffs in the 80s/90s ...
The US believed in free trade precisely when the politically connected needed labor arbitrage, and protectionism exactly when the politically connected needed protection. The pretense of underlying ideals was never more than a political tool - political economy was always political.
A real headscratcher isn't it? And from a government that is supposedly priding itself on small government. How should companies navigate this? What's the framework they should operate within?
Claiming the mantle of "small government" was simply an exercise in marketing to relax regulation meant to prevent bribery and corruption. In practice, the current slate of government officials believes in absolute control of whatever they want whenever they want.
It's a mirror case of the supposed "free speech absolutists" who immediately turned around and silenced, sued, fired or jailed once granted the power to do so.
This whole administration is absolutely rampant with corruption. Just yesterday we had JD Vance on TV saying that if Watergate happened today, it would just be a 12 hour news story, because they are getting away with so much worse.
Anyone who denies or defends this administration's corruption is complicit.
Munitions exporting. I fondly remember the PGP feasco. I spent years using PGP to encrypt my emails to several people who refused to use email without it. Good times.
Unprecedented? This is very much precedented and has been the end goal when you disallow regulations and public input when it comes to technology proliferation. When was the last time the public had an ability to direct technology in the US?
This is the result of private interests working authoritarian governments (hint, it rhymes with classism).
Yeah, that's what we need, frontier intelligence models open in the wild that, if a jailbreak is reliably established, there's no possibility for anyone to ever patch at the API layer. Because there is no API layer.
Between $5-10T of the US economy is subject to export controls. Nobody disputes that Mythos is dual-use technology, which means it has been export controlled since the day it was created.
Companies are responsible for demonstrating criteria to export (for example) a nerfed version (Fable) of an export controlled item (Mythos)
Nothing here is novel, unusual, capricious, or … fascisistic.
That threat will probably result in a DOJ investigation into everyone involved until they hit something they think they can prosecute someone for, even if it's not true.
Did you mean this in the French Revolution sense (the clergy, the nobility, and the commoners) or in the American sense (the legislative, judicial, and executive branches)?
The French Revolution sense would be an ironic counterpoint, because the Revolutionaries did end up capturing all three estates, only to fall to someone (Napoleon) who captured the military, which wasn't considered one of the "three estates" because at the start of the French Revolution destroying civil society, enacting a military dictatorship, and starting a series of wars throughout Europe was considered outside the Overton Window.
This perhaps holds some lessons for America today.
I think there is a layer of truth to the idea that MAGA captured what England would call the Fourth Estate, the News Media.
England is weird because its model looks like that French model, except that it intertwines the nobility and clergy. Basically an Archbishop and a Earl both sit in Parliament as Lords, while the Commoners control only the Commons of the Parliament. Now today the Commons runs things, but that's relatively modern, in the 19th Century it was completely normal for the Prime Minister to be a Lord, and while some of them were only technically Lords, having in fact been elected but just also nobility anyway for one reason or another, or being ennobled while serving as PM because nobody thought that was a bad idea - others were never elected at all.
So weirdly the place which came up with the "Fourth Estate" only really had two other estates, although everyone reading will have known about the French concept too.
In the era when Lords routinely become PM (it would still be legal to do this today, but it's hard to imagine it happening, although the Tories did give a Lord one of the Great Offices of State so never say never) almost all those Lords were born into it. Today basically nobody sat in the Lords was born to it (there are about two dozen left, when they die or retire that's the end of it) but there are still always a dozen or so bishops, and Iran is ironically the only other place [except the Vatican which barely counts] where religious leaders are in government by fiat in the modern world...
Edited to mention the Vatican before somebody else does.
But your government is constantly acting illegally. Isn't it time for Americans to... do something? It's clear that your legal framework isn't working.
Do what, exactly? Throw away our lives? 40% of the nation would rather die than vote for anyone other than a racist, and another 40% would tsk-tsk and say “that’s not how you’re supposed to do it”. There’s no revolution coming to save the day.
It’s a bit in general, because if you actually read the EU AI legislation, most of it follows the right ideas and provides more safety, in the sense that OpenAI and Anthropic used to pretend to care about, but never really did.
The ideas are debatable but generally correct. The EU's problem is that regulation stops at the ideas, and it is intentionally designed so to be impossible to ensure compliance in advance. So the regulation is really after the fact and a subjective judgment by regulators. So there's tons of risk even if you genuinely believe you're complying with the prescribed intents.
My opinion on EU regulation would flip 180 degrees if they offered any kind of pre-clearance where you could propose a product, feature, or policy and be told in advance if it meets their subjective requirements.
IMO you can have clear, specific requirements in advance, or you can have a body that provides interpretations of spirit-of-the-rules regulations in advance. Having neither is a problem.
(yes, I'm aware of the argument that if you tell companies what's legal in advance they will just do the bare minimum or find loopholes... I don't find that to be a legit rule of law system)
I understand that desire entirely, but I’m not sure if it would work that way. Take an ISO 27001 certification (or SOC, if you like): There is no one clear set of things to do, but both guidance and requirements that you need to address and be able to defend your concrete implementation.
And I generally like that a lot better than having a set of hard this-way-or-no-way checklists that invariably consist of 80% bullshit ceremony for giant corporations. ISO nudges you toward that too, but if you’re able to deliver the same security guarantees with less, auditors will usually be happy.
The same, in general, works for GDPR regulations as well: The law is mostly about doing the right things, but not spelling out the billions of cases and permutations and strategic decisions involving privacy in one way or another.
It's deliberately not prescriptive as the implementers are the ones best placed to solve for requirements - you don't want policy makers providing technical checklists. And it's not unstructured - ISO 42001 essentially encodes it.
Im not happy with the AI act in entirety either, but my point was that it’s hard to read it and say "this isn’t generally the right thing to do", where right means responsible and beneficial to society as a whole.
On some level though we have to be cognizant of the potential for harm these models have.
LLMs are still a little loosey goosey, and we are right on the cusp (if not there already) for an agent to hack a bank and steal money for some rando teenager with a penchant for jail breaking.
The regulations are and will be negative, but don't lose sight of what LLMs can do off the leash.
Too overbearing though and you get... Mistral? A continent that hasn't been on the leading edge of anything (other that expansion of the regulatory state) for decades, and Europe feels it in their employment numbers.
Current French (8.2%), Spanish (10.3%) or even Swedish (8.6%) unemployment would count as a disastrous recession in the US. In the US we call 2007-09 the "Great Recession", which peaked at 10.0%, and that relatively brief time left a generational mark. That's a somewhat routine number by EU standards.
Not to mention you end up with bizarre effects. If the UK were admitted as the 51st state it'd immediately be the poorest. (Yes, some EU countries are wealthy, but they're also the size of US counties, if we cherry pick just Manhattan we could make some spectacular comparisons too)
So, it's a complex issue but the tradeoffs are absolutely tangible yet often dismissed.
> Current French (8.2%), Spanish (10.3%) or even Swedish (8.6%) unemployment
This obviously doesn't tell the whole story, because it only measures people actively in the workforce. Meanwhile, a far larger portion of Sweden's population is actually employed compared to the US.
Sweden's laborforce participation rate is 76% and in the US it is 62%. Sweden's employment rate is 77% and US's is 59%. Which statistics are more important?
Bank should be more secure, if a random person with an LLM can hack them, they should have paid 100 random blue teamers with LLMs to hack them first to get more secure. Not AI's fault.
Pretty sure you mean red team here. While I've heard people refer to any offensive security (eg including blackhat) as 'red team' , it typically means people you've hired or contracted to try to break into your systems, whereas the blue team are people you've hired to build and operate your security defenses. Red and blue team are both your employees / contractors but perform different functions.
The purpose of policies like this is precisely to ensure that those 100 runs do happen first, rather than allowing a free-for-all where they have to race to secure their systems.
>but don't lose sight of what LLMs can do off the leash.
there is no such thing as an LLM "off the leash", it's not a dog, and even if it was a dog the person responsible is the owner. What is this bizarre attitude to a piece of software that makes people think existing laws don't apply?
If your LLM agent hacks a bank, you have hacked a bank, you will go to prison and that's entirely sufficient. People have been hacking banks for decades now, it didn't require the government to regulate C compilers and Emacs.
If your web browser hacks a bank, but you didn't know and didn't expect it to, have you hacked a bank? Why is an LLM different? What happened to mens rea?
Our AI czar, David Sacks, whined and moaned about the idea of regulation, even said Anthropic begging for some guidance was asking for “regulatory capture” and was gloating about how right he was they wanted it, 2 weeks ago.
No, it’s not leopards ate my face / irony / comeuppance, because that would involve regulation.
I understand it’s very satisfying if you wanted Anthropic “punished” for asking for real regulation to see this. I can’t deny there was a little bit of me at first that felt that way.
It’s untenable, a first order reaction, that I regret intellectually, because if you were against regulation, you’re certainly against waves whatever this is.
Your position is “he made it happen by saying he didn’t want it to happen” with maybe a side order of “when he knows Mr. Trump doesn’t like him”
I posit that these ideas are common, and come from a place of Mr. Trump is more or less a normal president because they all do bad stuff, and regulation is Creating Monopilies. To wit, there’s 0 reason to believe the person you originally replied to’s claim that Anthropic wanted to kill startups. It’s just a random implication of what bad regulation can do.
This applies to most things when it comes to the USG/citizens. Protectionism is communist unless they do it. Thinking about developing a nuke? Well bomb you first despite being the only people to ever use them. Free speech and press - unless we don’t like what you say.
They're not regulating though – they're arbitrarily blocking releases based on no clear criteria. The EU may be legalistic and rules-based, but I'd take that over capricious and arbitrary.
Yes, Trump is the worstPresident we’ve had, the Supreme Court is captured by conservatives, the Republicans would rather die than vote against Trump, we know all of this.
The EU is nothing but capricious and arbitrary. Much of the DMA and similar is pure vibes that you can't know if you violated until the regulators do their divinations months or years after you shipped.
It sounds nice but you end up with entrenched special interests that later oppose all regulation regardless of the consequences. We have pesticides you wouldn't want anywhere near your children casually used to control weeds on kid's playgrounds, insanely huge trucks that kill hundred each year, the food is garbage...the list is long and tiresome. Trust me brother, if I could live in the EU, I would.
I agree. But that need has absolutely arisen. The US government is not exactly the best steward for this kind of thing, but some model other than "race each other as fast as we can" is desperately needed here.
Let's plan a fire fighter division only once we are actually having some buildings in the city burning down. That people who fear that ridiculous perfectly controlled fire in chemines are ridiculous.
Well, when the leaders of this movement go around doom-trolling for years on end this is what happens. It turns out you need to be careful what you say if you're a highly visible public figure. Amazing!
Honestly, with open source models I don't think this regulation means anything because there's no way they can really regulate what's coming out of china. I don't think this affects innovation in AI much at all (unless your definition of innovation is "pour more money into diminishing scaling"). It's mostly just bad news for the US frontier labs, and based on their behavior I don't feel sad for them AT ALL. Like, they've basically alienated the vast majority of people by outright threatening their livelihoods or even society at large, and now we're supposed to feel sorry for them because they can't just go around saying "THIS WILL REPLACE ALL JOBS IN A MONTH!" without consequences?
> because there's no way they can really regulate what's coming out of china
Do you think the Chinese will go parading around that they've created the greatest cyberweapon known to man, and the CCP will be totally cool with the Americans being first in line to buy tokens, because hey, free market?
They would sooner put all their own employees in an incinerator than allow that to happen.
I don't see how, other than that it will make it harder for chinese labs to train their models on OpenAI/Anthropics' (which honestly I can't get that worked up about plagiarism in this space considering where they got their data from..)
I remember when Republicans told us they want less regulation and smaller federal government. Now they want their dementia riddled god king to control everything from pool liners to the information you're allowed to see, which is all in books and readily available online.
It will just mean US providers will rapidly loose their moat. Their moat is already shrinking. If they can't release their best models, it'll shrink a whole lot faster...
This is what OpenAI/Anthropic want, it's better marketing than they can pay for -- and it creates a precedent for permanently banning the next generation of open weights models
The problem is they've convinced mainstream people that a model that can find a bug in Microsoft Windows is a bigger problem than Microsoft not caring about fixing it.
I largely blame people like Amodei for such outcome. As product owners, they could've do it the old way: telling people how great this product is, how much potential it has, and what kind of guardrails the companies are building and etc. But oh no, Amodei has to do the doom trolling 24x7, while in the meantime plays a cult leader by telling people only he knows how to the guarding angel of the AI or the humanity thereof. Ironically, the same people also push their companies to develop more powerful AI in full speed. They think ordinary Americans are so stupid that they can't see through them?
The AI companies were all asking for the government to regulate them. The government is doing what the companies asked for them to do.
You can argue that, by government, they meant some legislative process, but I'd argue that regulation via bad executive order is much better than regulation via bad legislation because the former is tractable. I say this as an EO minimalist.
What are the proper laws and regulations? Can you point me to a proposed framework that you believe is most correct?
I have no idea how this stuff should be regulated. I do know that any sort of comprehensive legislation at this point in time has a much higher chance of being a bottleneck to innovation than an easily reversible white house directive.
Of all the terrible things to come from the odious Trump administration, them saying "hey, can we make sure these models aren't dangerous?" is one of the least bad things they've done.
>I have no idea how this stuff should be regulated.
That's why we have a system where representatives of districts do research, debate, and hash out those details while the public who votes for them is able to contact them.
>I do know that any sort of comprehensive legislation at this point in time has a much higher chance of being a bottleneck to innovation than an easily reversible white house directive.
That's odd to say after admitting you don't know what the regulation would look like. Especially after seeing the "easily reversible" tariffs from this White House, which changed erratically and had exceptions for people who sweet talked the president.
If you don’t see a difference between a well thought out & debated policy stance and arbitrary enforcement without justification I’m not sure what to say.
They can’t, though. The models might or might not be too dangerous but the people running the US federal government are too incompetent and/or corrupt to do anything useful about that.
Anthropic submitted a long, thoughtful framework proposal, which everyone seems to be ignoring in favor of hot takes like “they asked for this!” No they didn’t, not at all.
> The AI companies were all asking for the government to regulate them
pretty shallow take. they asked for sensible, transparent, tech aware regulations. this is not that.
Ok so to be clear you agree this has not been the norm. It seemed like you were clarifying your original message but it was a change of topic, from "this has been the norm" to "this hasn't been the norm but they got what they were asking for" (or what they deserved if not exactly what they asked for). I'll dip out of that conversation.
You moved the goalposts. The government controlling what openai can and cannot do is completely different than they gating access out of their own volition.
My comment was in response to the parent's original comment: "Ok so you agree this has not been the norm," which didn't give me much to respond to. It has been edited since.
Yeah pretty sure we had a whole bruhaha ~250 years ago about this question of where precisely power belonged. I for one think we mostly got it right then and would be reluctant to shift the power back to the individual sovereign and away from the people.
I agree that this is all ridiculously arbitrary right now, but it shouldn't be surprising either.
I can't find the exact blog post (maybe on simonwillison.net ?) but I read people predicting that know your customer laws would be coming to AI if it gets more powerful several years ago already.
Powerful technology that can do immense harm in the hand of individuals/small groups is the most obvious (and legitimate!) target for regulation. Maybe Anthropics hubbub around Mythos made all of this happen earlier than it would have, but it was going to happen (if the models are going to get as capable as valuations imply they will).
Further more, no one actually gets hurt if we start rolling these things out more slowly.
Rolling them out selectively according to the whims of an administration that disdains fair process, tears down the institutions that could potentially provide and legitimise reasonable regulation, etc... well, that's another topic.
I wonder what kind of scheme the administration is up to. The obvious play is a squeeze where OpenAI and Anthropic are forced to give parts of their company away, like Intel. But they could also be toying with the idea of limiting frontier AI access to companies that bend the knee, which would further cement their grasp on the tech industry.
I’m sure they are wondering just as much. I assume exhorting Anthropic/OpenAI for personal bribes, favorable government contracts with no restrictions and public acts of submission.
Fable didn't silently dumb it down. It printed a warning that it has detected a possible inappropriate topic and you are being switched back to Opus. I hate it, but it isn't dishonest.
The big companies want this. It's a moat for them, a way to keep competitors (especially overseas) out of the US market.
They might try to extend it to downloadable open weight models, but honestly they might not even bother with that. The goal is to keep people from competing on lucrative contracts or the hosting market.
How can you simultaneously be a bottleneck for innovation while being their largest customer, and pouring tons of money and resources into it to help accelerate development?
The startup-brained among us never learned the first rule of business which is to not fuck over your benefactor.
I'm going to get downvoted here, but all the E/acc people that loudly allocated for Trump, someone known for amassing power by any means necessary including strong arming industry should be publicly eating crow right now. This was something that was always in the cards when you vote for someone who only cares about himself.
All the tech CEOs had no qualms about groveling before Trump and licking his boots, so yeah I assume they must be 100% onboard with stuff like this as well
It's worrying that with no formal and transparent policy framework that the government will be picking winners and losers and stifling innovation.
The market will demand such a framework. I suspect that's the larger idea here, in that Amodei not only wants to be in the room when that framework is written, he wants to be at the head of the table.
He apparently wants it so badly he's willing to set back his own company's IPO to make it happen, given that there can be no pure-play AI IPOs until the regulatory picture is sorted out.
Checks from the major model providers will already be on their way to Congress, hand-delivered by the highest-paid lobbyists on K Street. Look for them to wake from their ent-like slumber tout suite to pass legislation that the courts can use for guidance.
What Trump is doing at the moment is, as usual, only a distraction.
for america yeah. for the world the only real risks are american, chinese or corporate dominance. thats why its important to support open models wherever they come from and smaller players like mistral in france or black forest in germany.
I'm honestly surprised there isn't more political outcry. The administration has a party affiliation that, typically, insists on free market principles and is against government overreach and regulation.
You would think that this government, attempting to puppeteer the most rapidly growing industry in the world, would have more people outraged.
Where are all of the people crying "Communist"? This is one of those moments where it is less of an overreaction.
Also applies to Chinese models. Give it 5 more months of US admin locking out US models and let's see what the market will look like for OoenAI and Anthropic IPO.
Their innovation relies on a huge amount of investment made under the assumption that they'll continue to be able to provide frontier models to a global audience. If it turns out the US government only lets them sell gimped models to non-citizens then they'll forfeit the whole global market to China and investors will flee like rats.
Unfortunately this is better than the status quo which is a totally unregulated disaster. These models are only getting more capable and cannot simply be released whenever OpenAI or Anthropic feel like the vibes are good enough. We don't let passengers fly on unvetted jumbo jets either and we prevent them from flying when they have problems.
> We don't let passengers fly on unvetted jumbo jets either and we prevent them from flying when they have problems.
This is Mr. Fart's Favorite Colors all over again. Our "vetting" process is not any more useful than the billion-dollar metal detector you can skip with a TSA Precheck. It arguably does not deter the most dangerous attacks even slightly. What happens when a mentally-ill pilot locks their copilot out of the cockpit? Well, we write off a crowd of passengers and then "vet" the next jet as a safe vehicle.
AI will be the same way. These "safety" measures are performative and do not even slightly address the actual threat surface of the technology. Arguably, it cannot even be done.
What knowledge or skills do you have to be hating on the certification process for airplanes?
It’s just getting ridiculous at this point. There are plenty of industries regulated and certified by national or international agencies. And no they don’t get to do what they want.
Second of all, OP never even said anything about no regulation - they specifically said they wanted transparency which is 100% valid and better than a world where the government baby proofs everything for you
Models are already censored - and who they are or aren't uncensored for has a lot of implications which are way worse
And the jets is a terrible example - you picked one of THE highest regulated industries where NOBODY has a problem with regulation
The biggest concern is identifying "who". If the US government says only US citizens can access a model, how do they enforce that. Anthropic and OpenAI will use Persona (a company funded by Peter Thiel) to verify user identity. Verifying your identity with a government ID and linking that to AI is the dream of a surveillance state. Agents running on your computer, accessing your internet accounts, access to your personal conversations with AI, and accessible by the government is just wild.
No, not really. The biggest concern is the US government claiming it can decide who gets to use a product and brazenly kneecapping competitive companies who do such things as "write contracts and comply with them."
> Only companies approved by the government will get access. There is no process for individual users to get access to the new model.
I knew the time would come when individuals on personal subscriptions get the short end of the stick. Didn't think it would come so soon. I hope we're not too badly deprecated in the months to come.
Looks like I've got to improve my DeepSeek workflows.
For personal use I don't care if I get access to it. Tokens are becoming too expensive. I am using Chinese models. What worries me is that my company may never get access. I work for a well known US company, but from Europe, we also have developers in Mexico. I can only guess US gov will take this into account when deciding who gets to use the new models.
Even worse than not getting access is getting fired. Since less than 20% of our developers reside outside US and our management is suffering from AI hype, they can decide to close foreign offices as a way to get access to new models.
> For personal use I don't care if I get access to it
There's a big difference between being priced out of a market option and the government saying you literally cannot buy it. We should all be wary of government controls like this.
Also I'd be willing to bet companies in left-leaning states, or with left-leaning figureheads will be excluded from use as well. This administration is amazingly petty and has captured voter records for a reason.
Given how the WH operates these days, this is ripe for corruption. Imagine the WH dislikes the CEO of a biotech company, while appreciating the attitude of a competitor CEO. What is to stop them from stalling on giving acess for the latest model to the company they don't like?
In most countries around the world corruption/bribes are necessary for doing business. Companies even account for it on their books. It was about time the US caught up.
The tragedy is the Trump admin is setting the precedent and creating the framework which will be abused in the future. For all the complains he made about the deep state, he's just creating a new avenue for them to abuse power
Same as it ever was. If you listen to interviews with Zuckerberg he’ll talk about the constant communiques from The White House during the Biden administration. Trump didn’t start this unfortunately he’s just more brazen about it.
Uhhh no. The White House (and other government agencies) have a well-established right to communicate with private entities. They do not have a right to coerce them. There's a blurry line between these, but not in Facebook's case, as they don't even claim they were coerced. Same with Twitter, whose lawyers testified under oath that Twitter's content moderation decisions were not coerced whatsoever.
Zuckerberg has specifically said they received requests from the government, they complied with some (as they have a right to do), declined others (as they have a right to do), none of which was under duress, and the response to non-compliance was expressions of "frustration" by the government officials.
By contrast, OpenAI's largest competitor just got kneecapped by the US government because they insisted that the US government comply with the contract terms the US government signed to literally months previously.
Well I was expecting it but if it's not going to become available in this subscription cycle for me I will cancel oai subs as I did with claude ones months ago...
moving to open weight models is trivial now, with optimizations and stuff glm 5.2 is roughly the same price as the best models around from multiple vendors.
unless I could atleast try and see Sol perform like 10x better I don't really have a reason to switch back.
I used Fable for like what 2-3 days at most and didn't really feel it was so much better, only difference was I had to prompt it less, not to get what I want but to get to a working output. Code quality was still shit, still made bad plans and analysis and so on.
This is the AI booster equivalent of "well it works on my machine." Works better for me != works great for everyone. I'm amazed how much people on HN seem to think that all coding is stupidly simple web apps.
In terms of startups, predicting tech, and all the things Hacker News is about, it mostly matters what the clever hacker can do, not whether the tool is ready for the mainstream.
If a clever hacker can get 10x results with an LLM, they're gonna outcompete the 90% that can't figure out how to replicate that result, and they'll be able to get about as much work done without that 90%
Factories, Agriculture, etc. - this is hardly the first time that pattern has played out.
I felt like it was about 10X better at "pretty" but straightforward 1 shot'ish type tasks. Not so different for complex and specific tasks in real code-bases.
Why do you say it was a lot better, what type of tasks were you testing it on?
> I felt like it was about 10X better at "pretty" but straightforward 1 shot'ish type tasks. Not so different for complex and specific tasks in real code-bases.
What metric are you using for "better" here? If I've got a straightforward task GPT 5.5 is going to 1shot it anyway.
Maybe not 10x; but it's fantastically talented at intuiting intent, reconstructing contexts, and making aligned judgement calls. You could throw Fable utter garbage and get great results. Fable felt like it was modeling me whereas gpt-5.5 is still very much is riding your prompt, your inputs.
I have bit of humility here as this is basically how I felt about 5.4->5.5; but 5.4 was very much a scalpel-very spiky weird intelligence. 5.5 sits somewhere in-between, but still spiky and verbose- code-maxxed; not a great orator, not a good proactive "here's the skip-connection you probably should be thinking about but don't seem to be weighing" in the way that Fable is. Fable is crisp.
Let's say we're prototyping an interactive tree (this is totally made up, but you get the idea):
Take this data input and convert it to a Sugiyama-style tree with hand-drawn feeling lines connecting the nodes. We need the ability to activate a random subset of nodes with a paint splash. The whole thing should feel organic, incorporating small motion effects. As the nodes are activated, the edges should look like a hand-drawn painting effect drawing toward the node, and then SPLASH onto the end node as it changes from black to deep red. The background should utilize a muted paper color, and we must adhere to this color palette for all elements (PALETTE).
...then back-and-forth 10 prompts or so to get the prototype I was looking for.
Comparing these types of things between Fable and Opus, something like this would be quite a bit less glitchy, prettier, and closer to the quick prototyping I needed than what I got with Opus 4.8.
Now, when I went deep into a complex codebase to fix a small issue or optimization that spanned many files and was fairly unique from anything in the training data, I didn't really see any noticeable difference between Fable and Opus 4.8.
Im not worried about this at all. The OpenAI, Anthropic and the US government can play this game all they want... They're just accelerating the development of open source models; and helping destroy the lead the US has built in AI, and their profit margins along with it.
This is like the battle between PostgreSQL and Oracle all over. Move up market, isolate yourself to enterprises, and watch while everyone else builds on PostgreSQL and erodes any technical advantage you had, until people just stop talking about you altogether.
The government will just claim that unsanctioned models have the potential to deliberately introduce security vulnerabilities when working on IT projects (e.g. be trained to strongly yet covertly favoring introducing compromised dependencies when you are not looking).
Then laws will be made to forbid organizations who use models other than those from the sanctioned labs to participate in critical projects on national security concerns.
All of a sudden, no business would risk using open source models anymore.
Then all they do is drive the usage of open models underground (copyright infringement is illegal too, and still common), stifle US companies, and accelerate the rest of the world decoupling from the US.
I hope they do it! It will have a positive long-term effect just like the Iran war footgun accelerates renewable energy transition.
That until it becomes illegal to have or use open source models without approval and licence from government. With more talks about on device scanning, this could be easily plumbed in. If OS detects there is open source model, it could brick your device or alert authorities. Then next step will be limiting what operating system you can install. Likely only those where you cannot remove client side scanning.
I wonder how you can reliably detect an open source model though. It can be stored in any binary format, and the weights can be modified slightly so that the float values are completely different while the network works the same. The binary that runs it can be obfuscated as well. Maybe the hardware could detect common LLM inference patterns at runtime? That would probably produce many false positives.
Eh probably easier ways to do this. Just sanction all entities that release open weight models for "illegal distillation". Enough to cross the risk threshold for most businesses in the west, and reduce future monetization opportunities.
It can't easily be plumbed in, though. I can spin up my own Linux build with none of that plumbing and do what I want with it. I can grab China's best models and use them or distilled versions on my own terms because OSS allows for that. Until hardware comes fully locked down and the models cannot be run on old hardware, both a long ways off, OSS is a way out.
In the not so distant future, all your coding edits (for work anyway) will be through centralized gateways. Think remote desktop environment where pasting from the client is disabled.
Reminds of the TikTok ban for security and safety only for it to be sold to a fellow crony. Can't help but see this play going down again. Threaten / Ban / Control / Pressure a technology+company, then get your cronies a seat at the board.
The cynic in me suspects they were salivating so much over the Spacex IPO they wanted a finger in anthropics 2026 IPO. Banning fable ~1 day after.
Are these models still relevant for people outside the US? I get the impression we're stuck on GPT 5.5 and Opus 4.8 pretty much permanently now, and relying on Chinese models in future.
If the US really cracks down on frontier model access, you'll see them make Chinese open models illegal. You might say "oh well, let them try", but they will just put direct and secondary sanctions on every company whose systems have used Chinese models in some way. They just have to make an example out of a few international companies and no one will dare to use Chinese frontier models, at least commercially.
in the extremely unlikely event that they do this, what will happen is that Chinese models will become "rebranded" with a wink and a nod by the token routers (at the very least, the non-US ones). there is a zero percent chance that corporations will not work around it if the models are good and cheap.
I don't think a critical mass of them will oppose the US. The most likely equilibrium is Chinese models being shut out of any US-aligned markets (i.e. Europe at the very least, also East Asia, etc.). Probably India, Russia, Brazil etc. will resist such pressure, but they are protectionist and resilient to trade wars anyway, at the expense of their own welfare of course.
Just because the US sanctions a country, doesn't mean the rest of the world needs to as well. As a Canadian we traded with Cuba even when the US had an embargo on them.
> but they will just put direct and secondary sanctions on every company whose systems have used Chinese models in some way
Yes, and the rest of the world would just nod worriedly and go along with it, at massive cost to their economies, rather than treating it like the protectionism it is and responding to it with crippling counter-sanctions.
You shouldn’t build a business that relies on any of these models. It’s a geopolitical and sovereignty risk now. Someone could just rug pull your entire stack.
Not only that, but using Opus 4.8 [1m] right now outside the US, and suddenly I only have a 500k context window. I really hope this is just a strange Claude Code bug, but I had access to a 1 Million window before, and it wouldn't entirely surprise me if context window length becomes another US export restriction.
The Anthropic page here seems to say that Max users should have access to the full 1 Million window for 4.8:
I was already setting up my infra to experiment with GLM 5.2 and its 1 Million token window before this happened. I think I'm glad I did.
EDIT: Found a solution, seems Claude Code 2.1.193 (or an earlier version I didn't notice) changed default settings, so that if you have Autocompact turned on it occurs at 50% of the context window. If you turn off Autocompact, the full 1 Million context window is restored. Another example of Claude Code quietly changing default settings sigh
You want to compact early though as sending the whole chat you will end up with a lot of tokens not in the cache which 1. Costs way more and 2. Will slow the request down as it has to process it all.
I do agree in cases where I'm using API and not the subscription, this would be very costly via API. Not sure why the tokens wouldn't be in the cache though? Seems everything should be cached as long as I'm within the 1 hour caching window? If I'm wrong about how token caching works, I'm eager to learn!
My other concern is, it isn't really a 1 Million context window if we can only use the first 500k, right? But now that I've found that I can re-enable it, I'm happy.
I've previously had sessions go to 700k tokens and still be okay, though it does start drifting at that 700k point. I'm regularly at 300k with no problem.
In the early days of the LLM era, there was lots of talk about how big incumbents, in particular google would be disadvantaged relative to “startups” like OpenAI because of their valuable legacy businesses that could be destroyed if something went wrong. Mainly people thought about big lawsuits but government action is similar.
Now OpenAI and Anthropic are big incumbents with Trillion dollar valuations at stake, so they can’t take any risks. Unlike google they don’t really have a thriving primary business to protect though, so without being able to continue to take risks and ignore regulation startup-style, it’s going to be a lot harder for them to stay relevant.
Pretty happy for anything that will throw some sand into the gears of AI development, given all the negative externalities that are becoming apparent, even if the admin is doing it for the usual dumbass reasons.
Not sure if the USG reserving all superior models for themselves and big corp is sand in AI dev? Inference is clearly constrained, people still want better models. Everyone will still use the best they can afford -- now additionally limited by what the USG allows them to pay for.
we still have this delusion that we can just pound our chests and throw our weight around. it's clearly not working, it's not helping the citizens, but people still demand this.
It's really interesting that someone can know all of these domestic Chinese names and yet declare the industry generally a "smoking ruin". Is it from a newsletter or something?
Because anyone who used these companies' products in China would see a pretty large ecosystem that's making a lot of money.
> It's really interesting that someone can know all of these domestic Chinese names and yet declare the industry generally a "smoking ruin".
They're different things. Just like you can be the most famous actor or singer and still be poor. Being popular, having good products and actually making money is not the same.
And it's all relative. Today if NASDAQ dropped 20% the world would declare it in ruins. Are the companies still "alive"? Yes.
> Because anyone who used these companies' products in China would see a pretty large ecosystem that's making a lot of money.
Not true. A lot of them e.g. the public listed 1s have reported increased competition and reduced margins.
If NASDAQ dropped 20%, it would have returned to the level last seen three months ago, in March 2026. Calling that "in ruins" would be a pretty big stretch.
To pick one example, OP talked about meituan like they haven't seen multiple meituan delivery riders per block every time they take a walk. It strains credibility. Why do they even know the name if they don't see that? Its not like they operate in America.
> Why do they even know the name if they don't see that?
We're in the age of the Internet. You don't have to be physically there to see anything. They could have just read it on the news, saw social media videos, etc?
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.
The US is now doing a softer and broader version of the same thing to trust-based export sectors. It’s not the same method but! it is the same mechanism. The main difference is that the US damage is more reputational than structural, so it could be reversed faster (only if policy stops telling customers that dependence on America is a political risk)
This was generated by AI and slightly re-written or end capped by a human. Anyone that knows who Meituan and Boss Zhipin are wouldn't make the claims this post makes. It's not reasonable that someone who could list off these companies and incidents would believe China's tech industry is in ruin. There's no way. This poster clearly promoted for the list, and wrote his commentary around it. Sad.
> defence - why would anyone buy US weapons after Greenland and Canada
Huh? US foreign military sales are up at all time highs
"Total exports by the United States, the world’s largest supplier of arms, increased by 27 per cent. This included a 217 per cent increase in US arms exports to Europe, according to new data published today by the Stockholm International Peace Research Institute (SIPRI)"
Orange and apples... China has very intentional policy behind those decisions. The US... Not so much. I don't buy that Trump and his whole cabinet are as dumb as they look, but they are only motivated by profit. And ignorance.
This was coming for a while. For years now there have been job postings for ai safety and not really what people expect. Jobs in places like RAND, funded off DOD grants, exploring the feasibility of building a bioweapon with off the shelf tooling and measuring how far along these tools are. Maybe they figured out it was too easy now, and this is the clamping down we are seeing in response.
What part? The job posts are factual. There are still some up on rand career website although the bio specific ones are all filled now. Here is their department page on the subject where they cover the scoping (1). Mirror biology seems like something out of sci fi but it is one of their main efforts it seems so the theory must hold some water. There's also concern about bioweaponry and pandemics. The rest is me connecting the dots.
Obviously the job posts were not what was in question. If you were "just connecting the dots" aka making stuff up then thanks that tells me all I need to know.
> Maybe they figured out it was too easy now, and this is the clamping down we are seeing in response
His post was from the start thinking out loud. It’s a solid contribution. It seems against the spirit of the forum here to respond to someone unauthoritatively thinking out loud with “you’re just making stuff up!”
How does this account for the Chinese models that are the ones people will use if they can't use OAI's or Anthropic's. Last time I checked, the US president doesn't have the ability to regulate the Chinese models. Considering this, do you still stand by your maybe?
Maybe they are looking into those too and a ban might be on the horizon. President makes their own rules now and controls the supreme court, you can't consider precedent anymore.
I really don't think this administration is capable of thinking strategically enough for that. I'm starting to think we lost the AI war about two weeks ago at 5:21.
> Organizations interested in model access may join the GPT 5.6 waitlist line, hosted at OpenAI's official Palm Spring satellite campus. Line begins at rear entrance with expedited VIP waitlist line options for holders of partnering cryptocurrency tokens. Application fee required for access to venue; waivers available for select US corporations.
what's weird, is my employees abroad (outside the US) have access to Anthropic Fable .... so what exactly did we prevent by limiting United States citizens from having access ....
Unfortunately we're not in a position where we can promise an exact date, but we expect it to take weeks (not days or months). It's the best coding model we've ever trained and we're bummed we can't release it to everyone yet. When we do launch, we'll share a lot more evals and testimonials and demos that help show what it's good/bad at. Personally hoping that both GPT-5.6 Sol and Fable 5 get broadly released soon so that everyone (myself included) can try them head to head.
You don't have to mention details, but is it internally a topic that your CEO hasn't even publicly criticized the Anthropic model freeze and are open ai folks seeing through the Musk/xai game that is in play here?
The US government is clearly not interested in security of the models - otherwise they would not have only frozen the latest model (that is only a bit better than the previous), but all - and not only for US citizens, but for all people.
Given what's happening in the US, I suspect Musk is trying to slow down both companies and damage their funding. His goal is pretty obviously to get an advantage over Anthropic and OpenAI.
This is the last wake up call for EU. After China starts controlling their models, in 5 years EU would be left with archaic technology compared to other major economies
The organizers of the No Kings rallies have done infinitely more to achieve political change in the United States than online commentators who make fun of them for being insufficiently cool and edgy. Effective activism is not about feeling superior to the normies who have families and are busy on weekdays.
They've convinced Democratic political leadership that they must fight Trump wholeheartedly, and convinced most of the country that Trump is a bad president who's failing to effectively address any problems. More change will be possible after the next election, when Trump's severe unpopularity causes the Trumpists lose control of Congress.
Measuring the influence of any mass protest movement is hard because they don’t have a formal role in the political system. We’ll never be able to know a counterfactual of what would have happened had they not taken place. If someone wants to take the position that No Kings, yellow vests, Occupy Wall Street, etc. were all equally pointless, I don’t think I have a strong counterargument to that.
What I see much more often is people who think only the latter two movements were good. And I’m not sure what heuristic could possibly get you there other than generalized opposition to normies.
This is a real head scratcher. Unless this is a very short term action it seems to have only downsides for everybody:
- people pay much more for US models than Chinese models because right now they're the best. Once they're no longer the best (since you don't get access to them) why would anyone pay several times as much for the same result?
- once you get a high amount of tokens flowing into China instead of US companies, they will train on those chats and their rate of improvement will only accelerate, making US models even less attractive over time
- the sky-high IPO are dead in the water, since their story of "we will replace a good chunk of all knowledge work in the world, capturing a few % of total global spend relating to it" turns into "we will make a bunch of money out of a few dozen S&P 500 paying for the best, and some pocket money out of whoever uses our overpriced models that are as good as Chinese models" - far less money overall. Losing access to untold billions of investor money certainly won't improve performance for the US labs
- all the non-US people start asking themselves why they're funneling money to US corporations who barely share any of the secret sauce compared to Chinese corporations who share plenty when it comes to LLM, including the models themselves (at least for now)
- Chinese models have significantly less guardrails, making for better end-user experience
- there is a small but non-zero chance Euros get off their asses and invest into AI, making something halfway decent and further fracturing the market which cuts into US profits
So what's the benefit here? I thought the Mythos situation was the current admin taking revenge on Antrophic for not kissing the ring, or simply looking for a bribe, but no matter which way I look at it it's a self-own. The only way this would make any sense is if AGI is imminent, which I don't think even the boosters are arguing at this point.
Theoretically US could outlaw Chinese models, but I'm not sure what it's supposed to accomplish as the rest of the world certainly won't, especially as long as they release open weights models that you can run without phoning home.
Yea not sure who would put money into an OpenAI or Anthropic IPO at this point. After this episode I promptly bought the hardware I needed to run local models (~$25k) and am extremely impressed. The price tag alone is worth the peace of mind knowing I will always have a locally running LLM that nobody can take away. I don't need Mythos-level intelligence to do 99% of my day to day work; Opus 4.8 was more than enough and I am pretty close to that with the open source models. So what happens when this hardware inevitably gets cheaper and cheaper and people realize they can run the models they actually need locally and without handing all their data over to these companies?
Only if you're doing cutting edge research or some highly, highly niche project would you need the frontier models.
> We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.
I'm very glad to see them say this explicitly and prominently.
You know what? Fine. This delays the OpenAI/Anthropic/etc hegemony and creates more space for local LLM adoption and development.
My company is very interested in local LLMs even just to cut back on codex spend. I imagine a lot of other businesses are, too. With the recent developments in open weight models, it seems like it's only a matter of time before they're frontier level, and any added delay in OpenAI and Anthropic models being publicly available is just more reason for businesses and individuals to try them out.
Just like the Iran war accelerating fossil fuel abandonment, this administration can't even do the wrong thing without fucking it up. I say we take this win.
"We believe in broad access, and we plan to make GPT‑5.6 Sol, Terra, and Luna generally available in the coming weeks. As part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch. At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly. During this preview, we will continue testing and coordinating closely with partners as we work toward broader availability. We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases."
This amount of courting the current administration is pretty scary imo.
> This amount of courting the current administration is pretty scary imo.
That’s ironic – I interpreted that paragraph with the opposite slant: positively. If that’s what the government mandates then these companies, in the end, have little choice, so was at least relieved to see them publicly pushing back.
They absolutely do have a choice, Anthropic and OpenAI could fight it in court. Iran showed Trump is a coward, he wouldn't risk tanking the only industry still keeping the stock market growing.
They did exactly that with supply chain risk designation, and look what it got them: the administration simply found another more effective way to punish them.
It’s all speculation but I think he would have no qualms about tanking the only industry keeping the stock market growing. But given Kushner’s OAI investments, Trump stands to benefit personally from not tanking the industry.
That's hardly a speculation that he cares very much about the stock market, more than any of his predecessors. It's also why he takes a L instead of going into an adventure in Tehran. Last but not least, "it's the economy, ..." is on everyone's mind, including his
“Wouldn’t it be a shame if we export controlled all of your models and revoked the visas and green cards of all of your non US researchers. You should really reconsider challenging our orders in court. Also remember you have 16% public support and if the president endorsed it a national data center moratorium would pass with bipartisan majorities.”
"Cool, cool, hey, what percentage of economic growth is directly attributable to the growth of our companies again? And thanks for revoking our researchers' permits, enjoy them helping out China!
Also, oops, looks like our model weights got leaked on 4chan. How unfortunate."
“President Trump’s strong and decisive action on endorsing the passage of this legislation prevented millions of jobs being lost and addresses the inflation ordinary Americans are facing from datacenters cornering the market for hardware. Thank you for your attention to this matter.”
I wonder what's going to happen when the administration rolls over to the OtherTeam(tm). If they've established a good relationship with Team A then Team B is automatically going to hate their guts.
seems pretty smart to me. opens doors and provides opportunities that those that don't court the government will miss out on. of course, if they're principled, that's okay (regardless of which admin it is), but the reality is most companies aren't. gotta get a leg up somehow.
I mean, it seems like common sense - a limited beta test before widespread rollout. I'm not convinced they'll ever come up with a good framework for dealing with the cyber & bio issues, but getting triggered by a beta test rollout seems overboard.
It is common sense, and with literally any other administration in the past century it would seem like a good idea.
I have zero confidence that this particular administration has any interest in regulating the industry for the good of the country, much less for the good of humanity. They will use regulation to maximize personal profit for themselves and their cronies, at the expense of the nation. I would not have thought that of any other US administration in the past 100 years.
In the longer run, it probably won't matter. If the level of corruption we see currently becomes the norm, then the US is facing much bigger problems than counter-productive industrial policy.
If you're arguing they should have believed the AI doomer hype years ago and developed decades of regs at the drop of a hat, sure, i guess you can. That's a topic for historians.
But, the question today is what to do today, a rolling deployment seems pretty hard to argue with.
The difference is that FDA approvals are a well-defined process with specific and actionable criteria for the release of a new product. Whereas this is the administration running on vibes and favouritism
I'm not going to defend the administration on most things, but your characterization isn't entirely fair. The record seems to suggest that the administration deferred to Amazon and the NSA, which seems sensible.
Perhaps you can fault them with not coming up with an objective framework earlier, but that's a different criticism.
150 years ago, Bayer Inc. was mass producing Heroine. 130 years ago Merck and Parke-Davis were mass producing Cocaine(TM) -- all with zero oversight. It would be another 50 years before we even had an FDA and another 50 before the FDA was a reasonably well-oiled machine with a solid set of processes and requirements. Even then, it couldn't really (and can't really today either) prevent these non-US companies (both Heroine and Cocaine were German) from making and selling elsewhere.
Anthropic's fear-mongering and marketing is the reason we have these restrictions in the first place.
Despite their virtue signaling, Anthropic is the only major lab that has never released an open weights model, has been caught intentionally nerfing a model after release (Opus 4.6), intentionally and silently degrades performance for suspected competitors and AI researchers, complains incessantly about distillation when everyone is doing it (and after they settled for pirating books), and wants to pull the ladder out from everyone trying to catch up.
They're anti-consumer and only concerned with holding the power themselves.
I'm not a fan of Altman, but Anthropic is the worst actor in the space, and I hope they lose.
Everyone in the space was talking about the automation of work from about day 2. People couldn’t stop themselves from talking about the way it was going to end work, and tech firms were firing people left right and center over AI.
Notably, Anthropic is the firm that stuck to its guns with the US Government, meaning they likely believe in their own spiel.
I mean it's fear-mongering until it isn't. I think people have become a bit too comfortable with dismissing the dangers of misaligned AI as simply "marketing hype".
What about openai's fear mongering, or googles, or JP Morgans, or Frank Herbert's, or Arthur C. Clarke's or Samuel Butler's?
If you can't envision plausible scenarios where very bad things happen because of a malevolent actor, ChatGPT 6, and a little bad luck - you need to think harder.
If the model is truly capable of causing harm in the wrong hands, then I think the restriction is quite reasonable. They'll patch the vulnerabilities and roll it out to the public in about a year.
> while AI companies and the administration work out a longer-term plan for regulation on the sector.
It’s not really the executives job or role to create new regulatory structures. If they want something durable, that lasts more than one administration, they need actual laws passed by Congress.
Last year Tim Cook gifted Trump a custom, one-of-a-kind glass plaque with a 24-karat gold base [1]. (Cook needed a policy outcome that would protect Apple's supply chain costs and avoid a costly 100% tariff on certain chips and components.)
You may have to make similar offerings if you want to use the latest version of ChatGPT.
OpenAI/Anthropic are begging to be restricted because it's great marketing and it creates a precedent to permanently ban open weights models. The problem is nobody in government believes in/cares about commodity pricing of truly open AI and how much it could help the world economy and prosperity.
Companies like Microsoft have been asleep at the wheel in terms of security for decades and now there's a model that can identify where they've been careless. That's not a "nuclear bomb level threat" or whatever Anthropic wants to call it, it's reckless carelessness by the existing companies.
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.
What are the odds this is going to become another avenue for grift - magically any companies the trump family invests in are going to get access. Any companies that aren't sufficiently 'loyal' to the regime will have to wait or may never get access.
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.
The geopolitical angle of all of this is interesting. Will countries, especially bigger players really just hope they'll get access to something so crucial from the US or China?
Probably the EU could pool together funds to create something competitive as being on the mercy of someone else isn't a pleasant place to be.
And I wouldn't get so used to the open models. Eventually, if they get good enough, the access to them will also get restricted.
Yes, to the surprise of some HNers, regulations can be good or bad. Just because there are people unhappy with current regulation doesn't automatically mean regulation shouldn't exist at all.
BTW this isn't an opinion on the availability of GPT 5.6. I couldn't care less about that.
Imagine if someone was lobbying for some reasonable regulation (we should regulate drugs, based around clinical trials) and then instead of a transparent system you get purely executive actions with little to no public justification (Trump declares all glp1s illegal no one knows why exactly)
Would you levy the same two quote criticism of the reasonable call for regulation?
Why would they? Today you have datacenters in EU offering Chinese and European models from Deepseek v4 Pro and Mistral Medium down to some Qwen 3.6 35B for effectively peanuts compared to Anthropic.
Basically the signal is that the total market for any US AI company is capped at however big the US market is. As non-US AI converges to Opus 4.8 level parity, whatever is still non-US consumer base shrinks towards zero.
It will be much harder for Chinese models to close the gap than it is to keep the historical 6-9 months behind. Their models' performance are heavily propped up on distillation runs. The capital going to their frontier labs is 10x-100x smaller than US frontier labs.
Local AI and open-weight models are becoming something to no longer ignore. I've started a community around this @tokenstead on X and tokenstead.ai YouTube and much more coming. DGX Spark on route, RTX 5090s and much more exciting builds. We need to have AI sovereignty!
> 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...)
I can't tell if this is bad for the big labs, or good because it means they now have an excuse for not showing meaningful progress in the lead ups to their IPOs.
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 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.
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.
The comparison to cryptography export restrictions is absolutely valid, but is that really favorable?
I'd argue that 70's cryptography export bans in hindsight look completely misguided, futile, burdensome and pointless in the end (which is why most of it was lifted/reverted over the last decades).
I don't see how AI-models are much different; it's certainly a better comparison than the fuzes, because we're both not at war right now and the underlying principle is already out of the bag.
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: billionaire 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 Fable 5.
And in the linked post, it says GPT-5.6 access decisions are supposedly just hand picked.
The stories about export controls are just songs they sing to the peasants.
There are claims that Chinese companies are mining + reselling Claude subscriptions like crazy anyway.
Because the models don't necessarily need to be hosted on hugging face.
You can create a Model Card repository containing your README and from there you include instructions or a custom script in your repository that allows authenticated users to download the model.
> Doesn't sound dystopian enough without a second compelled entity?
This is the second snarky question you've made today, the other in relation to the export limit.
> Is this just upsetting because it's a product you want to enjoy?
Both are assumptions you are making and don't provide much in the way of constructive conversation, if I'm wrong about something it's alright to just point it out.
> don't necessarily need to be hosted on hugging face.
Export restrictions don't split generally hairs on technicalities like "hosting" - the "but magnet links aren't actually torrents!" defense doesn't fly when $1M fines and federal felonies are at stake. All distribution or "causing" distribution to restricted entities is prohibited.
> This is the second snarky question you've made today
It's not snark: why would Cloudflare somehow be legally or technically relevant in the context of two American companies distributing export-restricted materials? HN seems to love the "Cloudflare controls the internet!" "NSA bad!" trope.
> Export restrictions don't split generally hairs on technicalities like "hosting" - the "but magnet links aren't actually torrents!" defense doesn't fly when $1M fines and federal felonies are at stake. All distribution or "causing" distribution to restricted entities is prohibited.
So why would open models that are not in the US be restricted ? The government would need to subpoena each model that was in the US individually, why would they do that when they could simply pull clout over CloudFlare, which we have seen governments do around the world. Either CloudFlare comply, or they're added a block list.
> So why would open models that are not in the US be restricted ?
Nobody said they would be?
> subpoena each model that was in the US individually
What does this even mean? Where did 'subpoenas' come into this conversation and how would that be useful?
> simply pull clout over CloudFlare
Cloudflare is an American CDN. Hugging Face is an American catalog/distributor (whatever semantic game you want to play) of models. Some of those models could be declared export-regulated. No subpoena is necessary to prevent Hugging Face or Cloudflare from distributing ITAR/EAR software, declaring any model as such, nor is trying to block something heavy-handedly at the CDN level necessary: Hugging Face will gladly comply with fine-grained requests.
"La Liga" obviously isn't American, which is why Spanish courts are compelling their ISPs (who they actually do control) to block Cloudflare IPs. Cloudflare's customers - who are likely not Spanish - are distributing materials Spanish courts do not approve of. If Spain had the means to compel Cloudflare or their customers in question to do anything, they wouldn't need to take such a blunt approach and block other legitimate customers. Cloudflare isn't involved in that equation and this isn't at all equivalent.
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.
This isn’t just about AI: they do not want you to privately use computers at all, in a way that cannot be surveilled. They want to extinguish all forms of general purpose computing and restrict you to walled gardens of apps and all code written via AI with the government-in-the-middle. It will be illegal to write code by hand, only people with “bad intentions” do that, because otherwise, why not just use AI like a sane person?
When I predicted this several months ago, here (mine my comment history) I was berated and downvoted, but primarily ignored. I have records, timestamped, of predicting this a year ago. Alright, great for me, pat myself on the back and get stuffed. Roger that.
What disturbs me is that this was not extremely obvious and predictable to everyone else. I have been called schizophrenic for my views on AI, here, and I kind of see how some could miss my points, but I am genuinely perplexed by the views on the subject I see around here, or specifically the views I don't see here.
Did anyone really not see this coming long ago? I have year-old transcripts discussing exactly what's happening with AI.
I started to have the opinion that the Chinese models would crash the AI bubble simply because they are an order of magnitude cheaper and almost as intelligent.
But if the government can simply ban models from the market? especially given how much the admin loves the idea of Tariffs? Knowing Trump the chance of this happening is 99.9%
We will all be stuck paying $50/mtok to Anthropic (And by we I mean only Big Tech will be able to afford tokens). The rest of the competition will be outcompeted by super intelligent machines. And AI CO’s /Big Tech will take over the economy.
Anthropic broke with US Gov over wanting restrictions and n how they use their model. OpenAI was more than happy to bend over backwards and hide behind a misleading press release.
The idea that OpenAI is the one who are meaningfully pushing back against the USGov is risible.
It's called ghosted, shadow banned, too sensitive / shitshow; Basically a multiplier of 0.1 is added to this post's ranking, or similar, and it will need thousands of upvote instead of hundreds to show.
It's commonly applied silently to posts that simply don't look good or become a nightmare to manage the narrative of. It's a healthy way to manage a community while looking transparent.
I think it's sucky and cheap, but at the same time it's also the best solution.
If they do it, they should mark the post as such as give an explanation. Otherwise you never know if uncle "Sam" called and asked it to be treated as such.
There's a huge difference between 'pro market' and 'buddies with some big businesses' and this administration is making it very clear, at least to those who would see.
it kinda seems like openai is doing this willinglyy and not challenging it. if they weren't doing this willingly, how would this be legal? has congress already passed a law giving the executive branch regulatory powers like this?
"A system of government marked by centralization of authority under a dictator, a capitalist economy subject to stringent governmental controls, violent suppression of the opposition, and typically a policy of belligerent nationalism and racism"
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.
It's good that we can be sure the policy will be fairly applied for the best of reasons and any donations for new ballrooms, ponds, jumbos etc immediately before access is granted will be entirely coincidental.
Yes. I could not possibly be more radicalized against the current administration, and I'm in favor of this. Future models will be even more dangerous than the current ones and we must build processes now to control their release when necessary. I don't like how informal this process is, and I absolutely despise the people running it, but I strongly prefer it to no process.
Those taking issue with the clear deference to the current U.S. administration would seemingly prefer it be the exact same degree of preemptive compliance and collaboration, just done behind closed doors as it was with the Biden administration. The sausage is apparently far more palatable when you only find out about the overreach, pressuring, implied threats, and censorship years later in House Judiciary Committees. Or even better if you don’t through use of NSL gag orders or implied threat of lawfare!
Uh, they spent hundreds of billions to get their flag ship products blocked by a democratic body with no possibility of recovering their capital except by concession of the people's representative?
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.
It seems like this was entirely caused by Dario (and Anthropic as a whole)? When you run around marketing something as a "super weapon", the government may actually take you seriously?
We obviously can't A/B test this... but if Dario hadn't been doing that, would any of this been happening right now?
Altman has done his fair share of "doom-trolling", claiming that his products are going to inevitably disrupt the global order in ways that demand government support and intervention. The entire industry has been marketing this way for years now.
> When you run around marketing something as a "super weapon"
That's one interpretation of what was said that ignores a lot of what was said.
So yes, if you ONLY read the headlines, sure. So, an ignorant and stupid government would read it that way. But the reality was, like many things, more nuanced.
However, I need not blame the messenger because the current government is led by idiotic morons.
Let's put this another way: either this is valid on behalf of the government, in which case he was right ot say something. Or you disagree with this, in which case, you can only blame the government for ignornig what was actually said.
What a party pooper the current US government is... I'm not excited right now at all, while normally a new GPT release would be so much fun to test out.
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.
> You wrongly assumed I implied these firms relocate to China. We are all aware of how China operates and controls its assets.
Then why write "+1 point to China!" and not "+1 point to the UK, France and Germany"?
> 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.
The UK, France and Germany all have their own export controls rules, so if a company in these countries comes up with a model that those governments deem to be of significant enough importance, they also have the means to exercise greater control over them as well.
The latest models from Anthropic and OpenAI are said to be the most advanced in the world. Agree or disagree, like it or not, the powers that be in the US determined that there is sufficient justification to control their export. Under long-standing and perfectly legal export control laws, the US has the ability to issue such orders.
In the case of Anthropic, the company chose to reverse providing public access to Fable because it said it could not comply with the requirement that non-US nationals (even those residing in the US) be restricted from accessing Fable.
> It’s slowing the entire country down at a very critical sink or swim inflection point in this tech.
You might or might not be right, but I think many people would argue that "move fast and break things" is risky when it comes to AI. I can't say that the current administration is genuinely concerned about the broad societal impacts of AI but if the effect of their brand of greater oversight is that companies like Anthropic and OpenAI have to slow down, it might not be a bad thing for humanity.
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
It's different, because most books don't contain the nuclear codes or have real impacts on national security.
The way I see so many comments on the internet hating any sort of AI regulation, is young juveniles cursing at the installation of stoplights as they rev their engines. The world is bigger than just you, and not only you matter. Reasons exist for doing things.
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.
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.
As a US citizen, I can purchase ITAR-regulated nightvision, IR lasers, etc.
But that's not what's happening. Frontier models are NOT being put under ITAR. Instead, they are being placed on an arbitrary "approved access" list. So that even if you qualify under export restrictions as a citizen, if you don't have a $200B+ market cap, you're disqualified.
Many people are upset about the national security restrictions, but it's MUCH WORSE than that. If I have to verify ID/citizenship, well, that sucks, but it would at least be an option. That's not what's happening here. If you are an individual or small business, no matter how "patriotic" you might be, you're out of luck.
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.
I'm not presuming, I flat out said: nobody sane would trust them with their business. They've been shown as unreliable suppliers due to arbitrary decisions by the White House. Nobody would want for their business automation processes to stop working because someone woke up pissy and banned the model they were using.
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.
This is where AI doomerism has taken us. I also hate LLM abuse, but pretending that they are going to destroy humanity has opened the door for eventual police state level control over computing. It’s hard enough fighting off the “think of the children” idiots, now we have to push back against hyperventilating technophobes who think the world is going to end unless we get computing under control. All while the political elites rub their hands in anticipation.
"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.
That's one way to look at it I suppose. The other is that these institutions had a tax break for a long time, not having to charge VAT like every other business. So I think quite a few people see it as a little unfair that the schools for rich kids get a tax break: and it is wealthier families that use private schools for the most part. It's not like these schools didn't know this rule change was coming.
I don't live in the UK these days, but one of the problems with the place is how complex the tax system is. All these little carve outs, sudden % cliffs, rebates and what have you. My first job was writing payroll software in the UK. You think that's the norm, then you move somewhere else and realize how much easier it is in many other countries. Then you get calls from people like "don't charge VAT on vegetables like in the UK": people don't understand the cost imposed administrating an ever more complex tax system.
It is extremely difficult for me to care about the fate of private schools. In my opinion, they shouldn't exist. If the rich are forced to send their children to the same schools as everyone else, maybe they'll pressure the government to improve said schools.
> We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.
My brother in Christ, then why did you (and your competitors) spend years telling the government you needed them to tie your hands behind your back? Did you really think they'd just give you a crown that says "Gatekeeper Of All Neural Networks"?
In the past few years, that's been primarily Anthropic, right? A lot of the really regulation-oriented people at OpenAI went to Anthropic, particularly after the failed attempt to oust Sam Altman as CEO (that was in late 2023).
brother from another mother here: I don't think they were begging for overreach from the executive branch, likely would have preferred legislation, especially the kind that could be molded by lobbyists.
Honestly, are people not getting what's going on? The US is turning into a personalist regime, there is no "government" per-se, there's a dude. There are no 'rules' there's only the dude's opinion and you'll do whatever he feels like today.
The way you know this is true is to imagine The Others in power. Sacks used to scream about government interference, but now that he's running (this part of) the government, obviously things are different.
The only constant is that David Sacks (& co) always believed he should have all the power.
Of course they know what’s going on, that’s why they voted for it again in 2024. They just think they will be in the group of winners that get some crumbs.
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% on 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.
Capitalism and Artificial Intelligence are the same thing. As AIs grow stronger, they will be turned upon themselves - RSI.
Nothing else matters.
So everything is right on schedule, it was long predicted that the general public will never be given access to powerful AI, because Capitalism needs AI for itself, so it can finally decouple from its current host, humans, and move into the next and final host - AI.
I have a feeling the latest incremental release of ChatGPT is maybe less threatening to the survival of all living species than a nuclear weapon, but I’m no expert so could be wrong.
Sure, but it’s a bit of a strawman. Define what you think is the appropriate level of risk to society without resorting to the extremes and it becomes a more productive conversation.
Is it just short of a model that can be used for bioterrorism? Security vulnerabilities that can cripple banks? Or just undermining the current capitalist incumbents? There’s a broad spectrum
I'd say something like: If you provide it to one company then you have to provide it to all companies. If you aren't comfortable with that, then either nationalize it or accept that no one gets it.
I can understand the sentiment, but that’s not how the govt has operated in the past. There are numerous examples of tech transfers to benefit a single company. It’s reasonably common in aerospace and health applications.
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.
Yeah but the real deal is talent; When enough people move around, this is no more 'sacred trace' knowledge. Plus, When you start with a known set of evals, there's really just a few to solve for.
The set of models solving really most used/solved problems is a known, as opposed to the cases where it's unknown, which declines with usage over time.
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."
A model trained on all the data X was trained on should be improved to the extent that X is already out of date. A model trained on X itself has all the errors of X and all of it's own. Society itself seems to show that model collapse is entirely possible today and was presumably a problem in the past given the significance placed on citation and going to original sources that predates obsession with credit.
Previewing GPT‑5.6 Sol: a next-generation model - https://news.ycombinator.com/item?id=48689028
It's worrying that with no formal and transparent policy framework that the government will be picking winners and losers and stifling innovation.
There's been no public policy, executive order, legislation, or otherwise on this, I wonder if anyone has filed FOIA requests for these decisions or the conversations between the Executive Branch and AI companies.
US never exactly believe in full on 'free trade'.
It's a mirror case of the supposed "free speech absolutists" who immediately turned around and silenced, sued, fired or jailed once granted the power to do so.
Anyone who denies or defends this administration's corruption is complicit.
Doubtful it'd hold in court; this admin would have to show that it's not corruption, because we'd all assume otherwise.
This is the result of private interests working authoritarian governments (hint, it rhymes with classism).
Companies are responsible for demonstrating criteria to export (for example) a nerfed version (Fable) of an export controlled item (Mythos)
Nothing here is novel, unusual, capricious, or … fascisistic.
No amount of rules can stop people who are willing to break them. Only enforcement can.
It’s a perfectly good system for government regulation.
The French Revolution sense would be an ironic counterpoint, because the Revolutionaries did end up capturing all three estates, only to fall to someone (Napoleon) who captured the military, which wasn't considered one of the "three estates" because at the start of the French Revolution destroying civil society, enacting a military dictatorship, and starting a series of wars throughout Europe was considered outside the Overton Window.
This perhaps holds some lessons for America today.
England is weird because its model looks like that French model, except that it intertwines the nobility and clergy. Basically an Archbishop and a Earl both sit in Parliament as Lords, while the Commoners control only the Commons of the Parliament. Now today the Commons runs things, but that's relatively modern, in the 19th Century it was completely normal for the Prime Minister to be a Lord, and while some of them were only technically Lords, having in fact been elected but just also nobility anyway for one reason or another, or being ennobled while serving as PM because nobody thought that was a bad idea - others were never elected at all.
So weirdly the place which came up with the "Fourth Estate" only really had two other estates, although everyone reading will have known about the French concept too.
In the era when Lords routinely become PM (it would still be legal to do this today, but it's hard to imagine it happening, although the Tories did give a Lord one of the Great Offices of State so never say never) almost all those Lords were born into it. Today basically nobody sat in the Lords was born to it (there are about two dozen left, when they die or retire that's the end of it) but there are still always a dozen or so bishops, and Iran is ironically the only other place [except the Vatican which barely counts] where religious leaders are in government by fiat in the modern world...
Edited to mention the Vatican before somebody else does.
It's N=1, but I believe both that the EU approach discourages investment and innovation in the EU, and that this US policy will do the same in the US.
My opinion on EU regulation would flip 180 degrees if they offered any kind of pre-clearance where you could propose a product, feature, or policy and be told in advance if it meets their subjective requirements.
IMO you can have clear, specific requirements in advance, or you can have a body that provides interpretations of spirit-of-the-rules regulations in advance. Having neither is a problem.
(yes, I'm aware of the argument that if you tell companies what's legal in advance they will just do the bare minimum or find loopholes... I don't find that to be a legit rule of law system)
And I generally like that a lot better than having a set of hard this-way-or-no-way checklists that invariably consist of 80% bullshit ceremony for giant corporations. ISO nudges you toward that too, but if you’re able to deliver the same security guarantees with less, auditors will usually be happy.
The same, in general, works for GDPR regulations as well: The law is mostly about doing the right things, but not spelling out the billions of cases and permutations and strategic decisions involving privacy in one way or another.
LLMs are still a little loosey goosey, and we are right on the cusp (if not there already) for an agent to hack a bank and steal money for some rando teenager with a penchant for jail breaking.
The regulations are and will be negative, but don't lose sight of what LLMs can do off the leash.
The appropriate level would be regulation though? Like I just don't get how we can argue that arbitrarily throttling companies is ok.
So here we are, it's probably going to me messy and err on the side of over-bearing.
Current French (8.2%), Spanish (10.3%) or even Swedish (8.6%) unemployment would count as a disastrous recession in the US. In the US we call 2007-09 the "Great Recession", which peaked at 10.0%, and that relatively brief time left a generational mark. That's a somewhat routine number by EU standards.
Not to mention you end up with bizarre effects. If the UK were admitted as the 51st state it'd immediately be the poorest. (Yes, some EU countries are wealthy, but they're also the size of US counties, if we cherry pick just Manhattan we could make some spectacular comparisons too)
So, it's a complex issue but the tradeoffs are absolutely tangible yet often dismissed.
This obviously doesn't tell the whole story, because it only measures people actively in the workforce. Meanwhile, a far larger portion of Sweden's population is actually employed compared to the US.
Sweden's laborforce participation rate is 76% and in the US it is 62%. Sweden's employment rate is 77% and US's is 59%. Which statistics are more important?
Pretty sure you mean red team here. While I've heard people refer to any offensive security (eg including blackhat) as 'red team' , it typically means people you've hired or contracted to try to break into your systems, whereas the blue team are people you've hired to build and operate your security defenses. Red and blue team are both your employees / contractors but perform different functions.
there is no such thing as an LLM "off the leash", it's not a dog, and even if it was a dog the person responsible is the owner. What is this bizarre attitude to a piece of software that makes people think existing laws don't apply?
If your LLM agent hacks a bank, you have hacked a bank, you will go to prison and that's entirely sufficient. People have been hacking banks for decades now, it didn't require the government to regulate C compilers and Emacs.
If your web browser hacks a bank, but you didn't know and didn't expect it to, have you hacked a bank? Why is an LLM different? What happened to mens rea?
I wonder if he understands why, now.
Anthropic was "begging" to make it harder for competing companies to be founded.
I understand it’s very satisfying if you wanted Anthropic “punished” for asking for real regulation to see this. I can’t deny there was a little bit of me at first that felt that way.
It’s untenable, a first order reaction, that I regret intellectually, because if you were against regulation, you’re certainly against waves whatever this is.
I would've much preferred if Dario hadn't run his mouth so much.
I posit that these ideas are common, and come from a place of Mr. Trump is more or less a normal president because they all do bad stuff, and regulation is Creating Monopilies. To wit, there’s 0 reason to believe the person you originally replied to’s claim that Anthropic wanted to kill startups. It’s just a random implication of what bad regulation can do.
I like the US approach better: regulate when the need for it arises, not before when you don’t know how the situation is going to evolve.
The fact that they couldn’t clear an already low bar is a really bad sign.
It’s bad, okay? And it’s not usually like this.
I agree. But that need has absolutely arisen. The US government is not exactly the best steward for this kind of thing, but some model other than "race each other as fast as we can" is desperately needed here.
Honestly, with open source models I don't think this regulation means anything because there's no way they can really regulate what's coming out of china. I don't think this affects innovation in AI much at all (unless your definition of innovation is "pour more money into diminishing scaling"). It's mostly just bad news for the US frontier labs, and based on their behavior I don't feel sad for them AT ALL. Like, they've basically alienated the vast majority of people by outright threatening their livelihoods or even society at large, and now we're supposed to feel sorry for them because they can't just go around saying "THIS WILL REPLACE ALL JOBS IN A MONTH!" without consequences?
Do you think the Chinese will go parading around that they've created the greatest cyberweapon known to man, and the CCP will be totally cool with the Americans being first in line to buy tokens, because hey, free market?
They would sooner put all their own employees in an incinerator than allow that to happen.
They were in the tank for Trump because they thought Biden/Harris would stifle them… and here we are.
open weight models - will be deemed too risky to be out in the open - since they can be abused by "bad actors" (unwashed masses)
ppl are acting like limited release is unprecedented when, in fact, has been the norm until a few years ago.
You can argue that, by government, they meant some legislative process, but I'd argue that regulation via bad executive order is much better than regulation via bad legislation because the former is tractable. I say this as an EO minimalist.
I have no idea how this stuff should be regulated. I do know that any sort of comprehensive legislation at this point in time has a much higher chance of being a bottleneck to innovation than an easily reversible white house directive.
Of all the terrible things to come from the odious Trump administration, them saying "hey, can we make sure these models aren't dangerous?" is one of the least bad things they've done.
That's why we have a system where representatives of districts do research, debate, and hash out those details while the public who votes for them is able to contact them.
>I do know that any sort of comprehensive legislation at this point in time has a much higher chance of being a bottleneck to innovation than an easily reversible white house directive.
That's odd to say after admitting you don't know what the regulation would look like. Especially after seeing the "easily reversible" tariffs from this White House, which changed erratically and had exceptions for people who sweet talked the president.
They can’t, though. The models might or might not be too dangerous but the people running the US federal government are too incompetent and/or corrupt to do anything useful about that.
I can't find the exact blog post (maybe on simonwillison.net ?) but I read people predicting that know your customer laws would be coming to AI if it gets more powerful several years ago already.
Powerful technology that can do immense harm in the hand of individuals/small groups is the most obvious (and legitimate!) target for regulation. Maybe Anthropics hubbub around Mythos made all of this happen earlier than it would have, but it was going to happen (if the models are going to get as capable as valuations imply they will).
Further more, no one actually gets hurt if we start rolling these things out more slowly.
Rolling them out selectively according to the whims of an administration that disdains fair process, tears down the institutions that could potentially provide and legitimise reasonable regulation, etc... well, that's another topic.
I’m sure they are wondering just as much. I assume exhorting Anthropic/OpenAI for personal bribes, favorable government contracts with no restrictions and public acts of submission.
It's protecting people from themselves, so basically like the safeguards already included in the models.
They might try to extend it to downloadable open weight models, but honestly they might not even bother with that. The goal is to keep people from competing on lucrative contracts or the hosting market.
The startup-brained among us never learned the first rule of business which is to not fuck over your benefactor.
The Project is almost here.
The market will demand such a framework. I suspect that's the larger idea here, in that Amodei not only wants to be in the room when that framework is written, he wants to be at the head of the table.
He apparently wants it so badly he's willing to set back his own company's IPO to make it happen, given that there can be no pure-play AI IPOs until the regulatory picture is sorted out.
What Trump is doing at the moment is, as usual, only a distraction.
We must clutch our pearls and cite National Security as a reason to pick winners and losers, just like the government did for Fable.
This is not something to joke about, its real.
You would think that this government, attempting to puppeteer the most rapidly growing industry in the world, would have more people outraged.
Where are all of the people crying "Communist"? This is one of those moments where it is less of an overreaction.
As predicted, [0] it has now been applied to OpenAI and soon anyone else releasing highly capable models.
[0] https://news.ycombinator.com/item?id=48511849
This is Mr. Fart's Favorite Colors all over again. Our "vetting" process is not any more useful than the billion-dollar metal detector you can skip with a TSA Precheck. It arguably does not deter the most dangerous attacks even slightly. What happens when a mentally-ill pilot locks their copilot out of the cockpit? Well, we write off a crowd of passengers and then "vet" the next jet as a safe vehicle.
AI will be the same way. These "safety" measures are performative and do not even slightly address the actual threat surface of the technology. Arguably, it cannot even be done.
It’s just getting ridiculous at this point. There are plenty of industries regulated and certified by national or international agencies. And no they don’t get to do what they want.
Unfortunately I have just as little trust in this instance of the US government as I do the corporations. Hopefully it's only two more years of this.
First of all, who said this is a disaster?
Second of all, OP never even said anything about no regulation - they specifically said they wanted transparency which is 100% valid and better than a world where the government baby proofs everything for you
Models are already censored - and who they are or aren't uncensored for has a lot of implications which are way worse
And the jets is a terrible example - you picked one of THE highest regulated industries where NOBODY has a problem with regulation
I'm hoping this is a call to action for local AI.
I knew the time would come when individuals on personal subscriptions get the short end of the stick. Didn't think it would come so soon. I hope we're not too badly deprecated in the months to come.
Looks like I've got to improve my DeepSeek workflows.
Even worse than not getting access is getting fired. Since less than 20% of our developers reside outside US and our management is suffering from AI hype, they can decide to close foreign offices as a way to get access to new models.
edit: grammar
There's a big difference between being priced out of a market option and the government saying you literally cannot buy it. We should all be wary of government controls like this.
there is no need to imagine, this is what is literally happening
You're two steps behind.
Zuckerberg has specifically said they received requests from the government, they complied with some (as they have a right to do), declined others (as they have a right to do), none of which was under duress, and the response to non-compliance was expressions of "frustration" by the government officials.
By contrast, OpenAI's largest competitor just got kneecapped by the US government because they insisted that the US government comply with the contract terms the US government signed to literally months previously.
moving to open weight models is trivial now, with optimizations and stuff glm 5.2 is roughly the same price as the best models around from multiple vendors.
unless I could atleast try and see Sol perform like 10x better I don't really have a reason to switch back.
I used Fable for like what 2-3 days at most and didn't really feel it was so much better, only difference was I had to prompt it less, not to get what I want but to get to a working output. Code quality was still shit, still made bad plans and analysis and so on.
It was a lot better. I can't believe people say this.
If a clever hacker can get 10x results with an LLM, they're gonna outcompete the 90% that can't figure out how to replicate that result, and they'll be able to get about as much work done without that 90%
Factories, Agriculture, etc. - this is hardly the first time that pattern has played out.
Why do you say it was a lot better, what type of tasks were you testing it on?
What metric are you using for "better" here? If I've got a straightforward task GPT 5.5 is going to 1shot it anyway.
Take this data input and convert it to a Sugiyama-style tree with hand-drawn feeling lines connecting the nodes. We need the ability to activate a random subset of nodes with a paint splash. The whole thing should feel organic, incorporating small motion effects. As the nodes are activated, the edges should look like a hand-drawn painting effect drawing toward the node, and then SPLASH onto the end node as it changes from black to deep red. The background should utilize a muted paper color, and we must adhere to this color palette for all elements (PALETTE).
...then back-and-forth 10 prompts or so to get the prototype I was looking for. Comparing these types of things between Fable and Opus, something like this would be quite a bit less glitchy, prettier, and closer to the quick prototyping I needed than what I got with Opus 4.8.
Now, when I went deep into a complex codebase to fix a small issue or optimization that spanned many files and was fairly unique from anything in the training data, I didn't really see any noticeable difference between Fable and Opus 4.8.
“Alice is supposedly smarter than Bob, but they both take the same time to tie their shoes.”
This is like the battle between PostgreSQL and Oracle all over. Move up market, isolate yourself to enterprises, and watch while everyone else builds on PostgreSQL and erodes any technical advantage you had, until people just stop talking about you altogether.
Then laws will be made to forbid organizations who use models other than those from the sanctioned labs to participate in critical projects on national security concerns.
All of a sudden, no business would risk using open source models anymore.
I hope they do it! It will have a positive long-term effect just like the Iran war footgun accelerates renewable energy transition.
Maintaining a blessed whitelist is the way to go.
lol that’s a good one.
The cynic in me suspects they were salivating so much over the Spacex IPO they wanted a finger in anthropics 2026 IPO. Banning fable ~1 day after.
Yes, and the rest of the world would just nod worriedly and go along with it, at massive cost to their economies, rather than treating it like the protectionism it is and responding to it with crippling counter-sanctions.
The excuse they give is borderline childish. I get the thing about slow rollout, make sure partners get to fix the bugs, etc...
But bad actors are hard working motivated entities with tens of thousand of fake ids, and american citizens working for them, for pennies.
All while the ones like or you sit at a crossfire which is borderline useless.
I cant wait to see what Qwen did with the massive distillation they made out of Opus 4.8 and Fable aka Mythos aka pretty sure they jailbroke it.
The Anthropic page here seems to say that Max users should have access to the full 1 Million window for 4.8:
https://support.claude.com/en/articles/8606394-how-large-is-...
I was already setting up my infra to experiment with GLM 5.2 and its 1 Million token window before this happened. I think I'm glad I did.
EDIT: Found a solution, seems Claude Code 2.1.193 (or an earlier version I didn't notice) changed default settings, so that if you have Autocompact turned on it occurs at 50% of the context window. If you turn off Autocompact, the full 1 Million context window is restored. Another example of Claude Code quietly changing default settings sigh
My other concern is, it isn't really a 1 Million context window if we can only use the first 500k, right? But now that I've found that I can re-enable it, I'm happy.
I've previously had sessions go to 700k tokens and still be okay, though it does start drifting at that 700k point. I'm regularly at 300k with no problem.
Big boxes with Huawei GPUs and Chinese open models to run inside your company without network access.
Now OpenAI and Anthropic are big incumbents with Trillion dollar valuations at stake, so they can’t take any risks. Unlike google they don’t really have a thriving primary business to protect though, so without being able to continue to take risks and ignore regulation startup-style, it’s going to be a lot harder for them to stay relevant.
It’s available to large companies. The WH gives them a competitive advantage against the rest of the market.
... that are friendly to this administration.
This will tank the market.
See you all on the other side!
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
Because anyone who used these companies' products in China would see a pretty large ecosystem that's making a lot of money.
They're different things. Just like you can be the most famous actor or singer and still be poor. Being popular, having good products and actually making money is not the same.
And it's all relative. Today if NASDAQ dropped 20% the world would declare it in ruins. Are the companies still "alive"? Yes.
> Because anyone who used these companies' products in China would see a pretty large ecosystem that's making a lot of money.
Not true. A lot of them e.g. the public listed 1s have reported increased competition and reduced margins.
If NASDAQ dropped 20%, it would have returned to the level last seen three months ago, in March 2026. Calling that "in ruins" would be a pretty big stretch.
We're in the age of the Internet. You don't have to be physically there to see anything. They could have just read it on the news, saw social media videos, etc?
Political priorities and good governance is why we have government.
Not really, no. What planet is this on?
- 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.
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...
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.
Huh? US foreign military sales are up at all time highs
"Total exports by the United States, the world’s largest supplier of arms, increased by 27 per cent. This included a 217 per cent increase in US arms exports to Europe, according to new data published today by the Stockholm International Peace Research Institute (SIPRI)"
[https://www.sipri.org/media/press-release/2026/global-arms-f...]
It's the ONLY one (almost) that are actively tested and verified in real battles.
1. https://www.rand.org/global-and-emerging-risks/centers/ai-se...
His post was from the start thinking out loud. It’s a solid contribution. It seems against the spirit of the forum here to respond to someone unauthoritatively thinking out loud with “you’re just making stuff up!”
/s, maybe
(I work at OpenAI.)
Given what's happening in the US, I suspect Musk is trying to slow down both companies and damage their funding. His goal is pretty obviously to get an advantage over Anthropic and OpenAI.
Hey OpenAI model go hack the new mythos for me.
Battle bots, oppression version.
What I see much more often is people who think only the latter two movements were good. And I’m not sure what heuristic could possibly get you there other than generalized opposition to normies.
I don't think they have done much real, if anything.
Want frontier intelligence? Better not defy the current administration, or your competitors will have access to a better model you could never use.
- people pay much more for US models than Chinese models because right now they're the best. Once they're no longer the best (since you don't get access to them) why would anyone pay several times as much for the same result?
- once you get a high amount of tokens flowing into China instead of US companies, they will train on those chats and their rate of improvement will only accelerate, making US models even less attractive over time
- the sky-high IPO are dead in the water, since their story of "we will replace a good chunk of all knowledge work in the world, capturing a few % of total global spend relating to it" turns into "we will make a bunch of money out of a few dozen S&P 500 paying for the best, and some pocket money out of whoever uses our overpriced models that are as good as Chinese models" - far less money overall. Losing access to untold billions of investor money certainly won't improve performance for the US labs
- all the non-US people start asking themselves why they're funneling money to US corporations who barely share any of the secret sauce compared to Chinese corporations who share plenty when it comes to LLM, including the models themselves (at least for now)
- Chinese models have significantly less guardrails, making for better end-user experience
- there is a small but non-zero chance Euros get off their asses and invest into AI, making something halfway decent and further fracturing the market which cuts into US profits
So what's the benefit here? I thought the Mythos situation was the current admin taking revenge on Antrophic for not kissing the ring, or simply looking for a bribe, but no matter which way I look at it it's a self-own. The only way this would make any sense is if AGI is imminent, which I don't think even the boosters are arguing at this point.
Theoretically US could outlaw Chinese models, but I'm not sure what it's supposed to accomplish as the rest of the world certainly won't, especially as long as they release open weights models that you can run without phoning home.
Only if you're doing cutting edge research or some highly, highly niche project would you need the frontier models.
I'm very glad to see them say this explicitly and prominently.
My company is very interested in local LLMs even just to cut back on codex spend. I imagine a lot of other businesses are, too. With the recent developments in open weight models, it seems like it's only a matter of time before they're frontier level, and any added delay in OpenAI and Anthropic models being publicly available is just more reason for businesses and individuals to try them out.
Just like the Iran war accelerating fossil fuel abandonment, this administration can't even do the wrong thing without fucking it up. I say we take this win.
This amount of courting the current administration is pretty scary imo.
That’s ironic – I interpreted that paragraph with the opposite slant: positively. If that’s what the government mandates then these companies, in the end, have little choice, so was at least relieved to see them publicly pushing back.
They absolutely do have a choice, Anthropic and OpenAI could fight it in court. Iran showed Trump is a coward, he wouldn't risk tanking the only industry still keeping the stock market growing.
They did exactly that with supply chain risk designation, and look what it got them: the administration simply found another more effective way to punish them.
Also, oops, looks like our model weights got leaked on 4chan. How unfortunate."
https://www.nytimes.com/2026/06/04/podcasts/the-daily/trump-...
I mean, it seems like common sense - a limited beta test before widespread rollout. I'm not convinced they'll ever come up with a good framework for dealing with the cyber & bio issues, but getting triggered by a beta test rollout seems overboard.
I have zero confidence that this particular administration has any interest in regulating the industry for the good of the country, much less for the good of humanity. They will use regulation to maximize personal profit for themselves and their cronies, at the expense of the nation. I would not have thought that of any other US administration in the past 100 years.
In the longer run, it probably won't matter. If the level of corruption we see currently becomes the norm, then the US is facing much bigger problems than counter-productive industrial policy.
But, the question today is what to do today, a rolling deployment seems pretty hard to argue with.
Perhaps you can fault them with not coming up with an objective framework earlier, but that's a different criticism.
Despite their virtue signaling, Anthropic is the only major lab that has never released an open weights model, has been caught intentionally nerfing a model after release (Opus 4.6), intentionally and silently degrades performance for suspected competitors and AI researchers, complains incessantly about distillation when everyone is doing it (and after they settled for pirating books), and wants to pull the ladder out from everyone trying to catch up.
They're anti-consumer and only concerned with holding the power themselves. I'm not a fan of Altman, but Anthropic is the worst actor in the space, and I hope they lose.
https://news.ycombinator.com/item?id=48465269
Everyone in the space was talking about the automation of work from about day 2. People couldn’t stop themselves from talking about the way it was going to end work, and tech firms were firing people left right and center over AI.
Notably, Anthropic is the firm that stuck to its guns with the US Government, meaning they likely believe in their own spiel.
I mean it's fear-mongering until it isn't. I think people have become a bit too comfortable with dismissing the dangers of misaligned AI as simply "marketing hype".
If you can't envision plausible scenarios where very bad things happen because of a malevolent actor, ChatGPT 6, and a little bad luck - you need to think harder.
It’s not really the executives job or role to create new regulatory structures. If they want something durable, that lasts more than one administration, they need actual laws passed by Congress.
You may have to make similar offerings if you want to use the latest version of ChatGPT.
[1] https://www.youtube.com/shorts/0O9QhwIkj5w
Companies like Microsoft have been asleep at the wheel in terms of security for decades and now there's a model that can identify where they've been careless. That's not a "nuclear bomb level threat" or whatever Anthropic wants to call it, it's reckless carelessness by the existing companies.
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.
Just because the many headed dragon is trying to bite your sailors' heads doesn't mean you should pilot your ship into the whirlpool
https://www.smithsonianmag.com/history/brief-history-book-bu...
"wait, not like that"
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.
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.
Probably the EU could pool together funds to create something competitive as being on the mercy of someone else isn't a pleasant place to be.
And I wouldn't get so used to the open models. Eventually, if they get good enough, the access to them will also get restricted.
BTW this isn't an opinion on the availability of GPT 5.6. I couldn't care less about that.
It's also more typical of a Reddit or YouTube comment, rather than HN, but that's a separate issue.
Would you levy the same two quote criticism of the reasonable call for regulation?
If they can’t freely sell access to their models and Chinese models catch up to Opus 4.8/GPT 5.5 in 6-8 months - then why pay OAI/Anthropic at all?
If non US citizens shouldn't have the models - 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.
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...)
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.
Any company providing the models will be deemed a threat to national security.
No need to block the download.
> 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.
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.
Countries are free to prevent exports of technology. Equating export controls with the Holocaust is disgusting.
I'd argue that 70's cryptography export bans in hindsight look completely misguided, futile, burdensome and pointless in the end (which is why most of it was lifted/reverted over the last decades).
I don't see how AI-models are much different; it's certainly a better comparison than the fuzes, because we're both not at war right now and the underlying principle is already out of the bag.
Just like every other export restriction on technology: once the actual cat is out of the actual bag, they are often relaxed.
The "underlying principles" here are hundreds of billions of dollars in R&D - which is what is required to compete with the frontier models.
> not at war right now
We weren't technically at war during the Cold War, either.
I'm more trying to invoke GRRM. This is a Game of Thrones: billionaire 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 Fable 5.
And in the linked post, it says GPT-5.6 access decisions are supposedly just hand picked.
The stories about export controls are just songs they sing to the peasants.
There are claims that Chinese companies are mining + reselling Claude subscriptions like crazy anyway.
Who is the "They" in "First They Came" referring to exactly?
> There are claims that Chinese companies are mining + reselling Claude subscriptions like crazy anyway.
Which will become a felony with export-controlled models, which is why identity verification is becoming a thing.
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?
> Doesn't sound dystopian enough without a second compelled entity?
This is the second snarky question you've made today, the other in relation to the export limit.
> Is this just upsetting because it's a product you want to enjoy?
Both are assumptions you are making and don't provide much in the way of constructive conversation, if I'm wrong about something it's alright to just point it out.
Export restrictions don't split generally hairs on technicalities like "hosting" - the "but magnet links aren't actually torrents!" defense doesn't fly when $1M fines and federal felonies are at stake. All distribution or "causing" distribution to restricted entities is prohibited.
> This is the second snarky question you've made today
It's not snark: why would Cloudflare somehow be legally or technically relevant in the context of two American companies distributing export-restricted materials? HN seems to love the "Cloudflare controls the internet!" "NSA bad!" trope.
So why would open models that are not in the US be restricted ? The government would need to subpoena each model that was in the US individually, why would they do that when they could simply pull clout over CloudFlare, which we have seen governments do around the world. Either CloudFlare comply, or they're added a block list.
> https://cybersecurityadvisors.network/2025/04/15/la-liga-blo...
This is not a new thing, anyway this discussion has become too argumentative for an off the cuff comment about government over-reach.
Nobody said they would be?
> subpoena each model that was in the US individually
What does this even mean? Where did 'subpoenas' come into this conversation and how would that be useful?
> simply pull clout over CloudFlare
Cloudflare is an American CDN. Hugging Face is an American catalog/distributor (whatever semantic game you want to play) of models. Some of those models could be declared export-regulated. No subpoena is necessary to prevent Hugging Face or Cloudflare from distributing ITAR/EAR software, declaring any model as such, nor is trying to block something heavy-handedly at the CDN level necessary: Hugging Face will gladly comply with fine-grained requests.
"La Liga" obviously isn't American, which is why Spanish courts are compelling their ISPs (who they actually do control) to block Cloudflare IPs. Cloudflare's customers - who are likely not Spanish - are distributing materials Spanish courts do not approve of. If Spain had the means to compel Cloudflare or their customers in question to do anything, they wouldn't need to take such a blunt approach and block other legitimate customers. Cloudflare isn't involved in that equation and this isn't at all equivalent.
wondering when more people will raise their voice and get engaged
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.
What disturbs me is that this was not extremely obvious and predictable to everyone else. I have been called schizophrenic for my views on AI, here, and I kind of see how some could miss my points, but I am genuinely perplexed by the views on the subject I see around here, or specifically the views I don't see here.
Did anyone really not see this coming long ago? I have year-old transcripts discussing exactly what's happening with AI.
But if the government can simply ban models from the market? especially given how much the admin loves the idea of Tariffs? Knowing Trump the chance of this happening is 99.9%
We will all be stuck paying $50/mtok to Anthropic (And by we I mean only Big Tech will be able to afford tokens). The rest of the competition will be outcompeted by super intelligent machines. And AI CO’s /Big Tech will take over the economy.
Really glad to see some reasonably prominent pushback against this government overreach.
The information has been reporting that the government wants to individually approve which companies get access and when.
Imagine the wonderful opportunities for corruption and influence peddling, not to mention, excluding any companies that don’t support Trump
The idea that OpenAI is the one who are meaningfully pushing back against the USGov is risible.
It's commonly applied silently to posts that simply don't look good or become a nightmare to manage the narrative of. It's a healthy way to manage a community while looking transparent.
I think it's sucky and cheap, but at the same time it's also the best solution.
https://blog.supplysideliberal.com/post/47857230937/luigi-zi...
It's a matter of time before the Chinese models are banned.
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.
This is rich social classes claiming more for themselves.
Someone convince me otherwise?
They are playing status games, and making particular products available to specific people.
Whatever the majority of people get will be a modified, probably weakened, version of what those at higher social classes get.
> with no possibility of recovering their capital except by concession of the people's representative
This is definitely not true and its not that binary.
Thank you Chinese Robin Hoods
I hope the country doesn’t become the new USSR.
- instant, total world war if it's not coming from USA
- let's finish all oil's reserve first otherwise
Censorship. Surveillance. (Hi, PLTR!)
We obviously can't A/B test this... but if Dario hadn't been doing that, would any of this been happening right now?
No it doesn't.
> When you run around marketing something as a "super weapon"
That's one interpretation of what was said that ignores a lot of what was said.
So yes, if you ONLY read the headlines, sure. So, an ignorant and stupid government would read it that way. But the reality was, like many things, more nuanced.
However, I need not blame the messenger because the current government is led by idiotic morons.
Let's put this another way: either this is valid on behalf of the government, in which case he was right ot say something. Or you disagree with this, in which case, you can only blame the government for ignornig what was actually said.
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 misspelled “nationalize them” (while we privatize Social Security and probably Medicare)
Only the bad parts of capitalism, only the bad parts of socialism. This is what policy looks like in the 2020s.
GLM on LLM Asics is going to be amazing, US hosted or otherwise
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.
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.
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.
Then why write "+1 point to China!" and not "+1 point to the UK, France and Germany"?
> 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.
The UK, France and Germany all have their own export controls rules, so if a company in these countries comes up with a model that those governments deem to be of significant enough importance, they also have the means to exercise greater control over them as well.
The latest models from Anthropic and OpenAI are said to be the most advanced in the world. Agree or disagree, like it or not, the powers that be in the US determined that there is sufficient justification to control their export. Under long-standing and perfectly legal export control laws, the US has the ability to issue such orders.
In the case of Anthropic, the company chose to reverse providing public access to Fable because it said it could not comply with the requirement that non-US nationals (even those residing in the US) be restricted from accessing Fable.
> It’s slowing the entire country down at a very critical sink or swim inflection point in this tech.
You might or might not be right, but I think many people would argue that "move fast and break things" is risky when it comes to AI. I can't say that the current administration is genuinely concerned about the broad societal impacts of AI but if the effect of their brand of greater oversight is that companies like Anthropic and OpenAI have to slow down, it might not be a bad thing for humanity.
-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.
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.
We’re still in the “5$ airport Ubers thanks to VC money” era of AI
They literally asked for this.
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.
These models aren't even that smart and they are already trying to control them and lock them down to a handful of people.
Then these executive and VC wonder why people hate AI and are against them.
Because the future is heading toward intelligence for the rich and you stuck with whatever model they want you to have.
The next step is banning open source models.
The future is not looking so bright if these models are already going locked down to whoever the government what's to have them.
This is no different than the government banning books because they don't want you to learn.
The way I see so many comments on the internet hating any sort of AI regulation, is young juveniles cursing at the installation of stoplights as they rev their engines. The world is bigger than just you, and not only you matter. Reasons exist for doing things.
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?
Except that they are.
As a US citizen, I can purchase ITAR-regulated nightvision, IR lasers, etc.
But that's not what's happening. Frontier models are NOT being put under ITAR. Instead, they are being placed on an arbitrary "approved access" list. So that even if you qualify under export restrictions as a citizen, if you don't have a $200B+ market cap, you're disqualified.
Many people are upset about the national security restrictions, but it's MUCH WORSE than that. If I have to verify ID/citizenship, well, that sucks, but it would at least be an option. That's not what's happening here. If you are an individual or small business, no matter how "patriotic" you might be, you're out of luck.
Did you read the E.O., or just Huffpo's interpretation?
> ITAR
This is more likely to fall under EAR, it's important to be aware-of and learn the difference.
> placed on an arbitrary "approved access" list.
Except that's not what the original E.O. indicated, this is just what Anthropic is choosing to do.
This kind of zero-sum thinking is what is killing the US's global influence right now.
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.
Good luck with "if you don't let us use your AI technology, we wont allow iPhones in" - go for it.
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.
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.
OpenAI Leans Toward Waiting Until Next Year for IPO
https://news.ycombinator.com/item?id=48678873
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.
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 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.
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.
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.
> 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
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.
Is that a feature or a bug?
That's one way to look at it I suppose. The other is that these institutions had a tax break for a long time, not having to charge VAT like every other business. So I think quite a few people see it as a little unfair that the schools for rich kids get a tax break: and it is wealthier families that use private schools for the most part. It's not like these schools didn't know this rule change was coming.
I don't live in the UK these days, but one of the problems with the place is how complex the tax system is. All these little carve outs, sudden % cliffs, rebates and what have you. My first job was writing payroll software in the UK. You think that's the norm, then you move somewhere else and realize how much easier it is in many other countries. Then you get calls from people like "don't charge VAT on vegetables like in the UK": people don't understand the cost imposed administrating an ever more complex tax system.
My brother in Christ, then why did you (and your competitors) spend years telling the government you needed them to tie your hands behind your back? Did you really think they'd just give you a crown that says "Gatekeeper Of All Neural Networks"?
[dupe] Earlier discussion: https://news.ycombinator.com/item?id=48678789
The way you know this is true is to imagine The Others in power. Sacks used to scream about government interference, but now that he's running (this part of) the government, obviously things are different.
The only constant is that David Sacks (& co) always believed he should have all the power.
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.
https://digital-strategy.ec.europa.eu/en/news/commission-sel...
A consortium will train a 400B-class model and get 2.5% on 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.
Nothing else matters.
So everything is right on schedule, it was long predicted that the general public will never be given access to powerful AI, because Capitalism needs AI for itself, so it can finally decouple from its current host, humans, and move into the next and final host - AI.
Just now, we're seeing what capitalist policies do to another nation, and that nation is us.
One can market socialism.
One can have crony capitalism.
Is it just short of a model that can be used for bioterrorism? Security vulnerabilities that can cripple banks? Or just undermining the current capitalist incumbents? There’s a broad spectrum
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.
The set of models solving really most used/solved problems is a known, as opposed to the cases where it's unknown, which declines with usage over time.
If anything I bet these people will just use their knowledge to make even more money reselling tokens.
People have no idea and everybody pretends to be an expert and ignore how good China is on AI research
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."