There's two basic kinds of distillation: 1) the massive [and dumb] method where you ask a question and use the answer as reinforcement (Black Box), and 2) more targeted distillation where you use one model to directly inform/train/guide another model (RLAIF).
The latter is basically fine-tuning the model with direction from another model. Thousands of businesses do this every day to fine-tune. This is almost certainly what the Chinese labs are doing, since it has a much better effect on the end result than just getting simple answers to simple questions.
These complaints of distillation are inflating the problem to make it sound worse than it is, because they want the USG to block/ban Chinese model providers as protectionism. They have already called for more export controls on chips (which is funny because DeepSeek v4 was designed to run on Huawei chips and now the other Chinese providers are following suit). But they can't come right out and say that, so their claim is that they're asking for more export controls because distilled models might not be as safe as their own. But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
> These complaints of distillation are inflating the problem to make it sound worse than it is
Unfortunately, the Reuters piece itself is complicit in this dramatization. The lede paragraph parrots Anthropic's talking point that distillation is an "attack", without using quotes that would alert the reader that this framing is a corporate talking point. Distillation is NOT an attack.
Stupid question: I was under the impression that these models were trained on PB of data. Surely the amount of questions/response they can extract from querying a bigger model (Claude) is fairly modest. How is it not a drop vs the training dataset?
From what I understand, at this point, the main value of stronger model outputs is simply to bootstrap reasoning behavior during the RL post-training phase. It gets you past the “cold start” problem with RL, after which the outputs aren’t needed anymore. From then on, it’s hill climbing and that requires environments for the model to interact with get rewards from.
> But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
They claim two things:
1) The specific, available jailbreak for Fable 5 is not dangerous - this has been confirmed by multiple experts, and there is no credible evidence against this claim (in other words, Anthropic is probably correct)
2) It is impossible to build an LLM that is immune to all jailbreaks. Again, there is no credible evidence against this claim, i.e. Anthropic is again entirely correct.
If #1 was false, they could just publish the details of the jailbreak - it supposedly only works on Fable 5, so there's no possible danger.
If #2 was false, surely some other LLM lab would have done it by now. Especially since a number of governments have made it clear there is a market for such a project.
Supposedly the details of the ‘jailbreak’ are that you give it insecure code and say “fix this code”, and it does, and then you ask it for test scripts and that’s effectively an exploit against the unfixed code.
https://research.nvidia.com/labs/lpr/slm-agents/ - Distillation data is a natural byproduct of using these models. There's no effective defence against it. Anthropic is degrading thinking blocks to summaries to slow it down and hide model internals, but in the end, the math says you're SOL and it works on MNC/Large Corporate scale well enough that the moment cost becomes a priority, you're left without the lock in you need to keep customers paying.
Similarly, if you did a corpus study on bioRvix to summarize recent science findings — you could use the same questions and answers to fine tune a model.
There is no way to communicate information at scale to companies through the API, for anything approaching a real application, without that information forming a corpus another model can be trained on.
But it wouldn’t be the first time they broke a model:
Their “guardrails” that cause it to reject user prompts also means it relies on its pop science summary of medicine to tell you why bioRxiv is wrong rather than accurately summarize the papers.
They’ve successfully created a smug, argumentative average of the internet which refuses to even consider it might be wrong or that it’s reading a science paper which is based on measurements and not vibes — but why would I pay for that?
> These complaints of distillation are inflating the problem to make it sound worse than it is
This is, in part, a problem every judicial and legislative system has faced since forever: form versus function.
Take a classic elicitation spying techniques: a foreign spy meets a military officer/scientist at a bar, strikes up a conversation, makes an observation wondering how could a missile hit some target at some accuracy and elicits a response that with laser guidance it is entirely possible. From this they get info that there is some technology to laser guide missiles. Or in retail, a competitor hiring a secret buyer for core baskets of goods and analyzing prices in the receipts.
The function is espionage, the form is conversation and all info is in a sense provided willingly. Where do you pull the slider?
These distillation "attacks" are not only indistinguishable from evals, they ARE evals. The function is own model training, the form is eval. Normally, one would expect to have risk benefit analysis based discussion which direction to push the legality slider to. The problem with these recurring statements is that they invoke enshitification of legislature.
I've used RLAIF to build out heuristic based non-LLM models for various decision systems and achieved like, 95% F1 on certain projects. We're in a place where models can be used to fine tune a lot of stuff via loops.
>But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
Yes this is in line with what Anthropic said in their public statements about their Fable access restriction by the government directive. The hypocrisy and inconsistency in their statements and behavior feels quite childish and controlling. I believe our companies and their leaders, friends among our other influential leaders and leaders from rich social classes, want to actively hurt most people as this behavior looks to be quite self-interested.
Not to mention, the person who brought this quote unquote jailbreak to the Trump Administration was Amazon’s new CEO. They know their IPOs are coming up, so locking their competitors out of the U.S. (even if just for the weeks surrounding the IPO date) would be a major boon.
The White House seems to love making announcements just for the sake of making the market move…. Coincidentally, right after POTUS buys a massive amount of the benefactory company’s stock
(Buy Dell Computers, lol)
> These complaints of distillation are inflating the problem
They’re also missing the point. What would have happened to a member of the Manhattan Project who, through personal pursuit of profit, neglected their duty enough to let the bomb leak?
You can do things like that - one example is averaging weights between related models - but not with Anthropic's models, because outsiders don't have access to the weights.
Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices. They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs. They are subsidizing model access in exchange for user logs and reasoning traces, which they then sell as training data, allowing them to operate below cost.
Claude and ChatGPT are both blocked in China. You need to use a VPN to access either, and you can't pay with a Chinese bank card. So most people who want access to Claude buy access via a reseller. It's the easiest and cheapest way to access Anthropic models in China.
These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
China aren't offering a cheaper solution. They are subsidizing an existing one (which is already subsidized) in order to gain foothold. The difference is that in the US subsidies come from VC, while OP implies subsidies come from the AI labs that buy the training data (which may as well also be VC backed, so just one extra hop).
This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market. As with most venture capital startups. I believe this VC tactic is a well documented "cheat code" to bypass market forces and build a monopoly. I find it hard to compare that with a free market.
However, I don't really mind China "stealing" from Anthropic. For us consumers we are getting the cake and eating it too. I.e we are getting rapid improvement to the tune of over a hundred billion dollars in funding, yet the market remains big enough that there's a chance of it not ending up as a monopoly in 20 years. And venture capital are footing the bill. A part of their investment is practically being redirected to fund Chinese AI development. It lets us live out our lives as happy CAC farmers[1].
So I would argue its not as much of a "cheaper solution" as it is intentionally and maliciously abusing another company's product to extract more value than the billing plans intend (given an average user), and further subsidizing the product by selling this data to competitors. But I don't necessarily think its a bad thing for us end-users. Nor for the market. But it is bad for Anthropic and its investors.
> China aren't offering a cheaper solution. They are subsidizing an existing one
Chinese labs are also pursuing legit frontier-advancing R&D into efficiency and publishing papers in the open, a culture that's in retreat at top American AI labs
> China aren't offering a cheaper solution. They are subsidizing an existing one (which is already subsidized) in order to gain foothold.
In my economics classes, we were told that (in a "free market" argument) the best thing to do if a subsidy is making something you want cheaper is to use it. You're getting your thing, and at a reduced cost.
(I'm not really replying to you per se, I'm curious how "free market" folks in these comments would respond to this.)
This is why I don't understand why people complain about impractically cheap Chinese solar panels. The rest of the world should buy enormous quantities and bankrupt the mofos and hugely benefit along the way. Then later they can set up their own solar panel industries.
Because they arent selling at a loss. The business pipeline is subsidized by the state. But end to end from mining the minerals to shipping you the solar panel everything is "in house". Its all in China. Thats why why can sell so cheap. Its even cheaper to make.
This narative that the CCP is just subsidizing all business to "beat america" is just dumb. Its the build process being made cheaper by the government. Not the final product.
As far as I can tell, the Chinese government itself is complaining about 'oversupply' in the solar panel market. Ie it doesn't sound like they are subsidising it anymore.
> China aren't offering a cheaper solution. They are subsidizing an existing one
So basically like US companies subsidizing offerings with selling user data, ads for crypto scams, manipulation for elections, making people addicted to gambling and so on?
Seems fair and an improvement as you can choose between that and not. Unlike say offerings from Meta where the data selling and efforts to further gambling addiction is always included.
> This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market. As with most venture capital startups. I believe this VC tactic is a well documented "cheat code" to bypass market forces and build a monopoly. I find it hard to compare that with a free market.
Why? Lots of people try this tactic, but hardly anyone ever succeeds. Meanwhile, the customer benefits.
Which part are we supposed to have an issue with?
The selling data to offer cheaper compute?
Products taking over markets with below cost pricing because they have money and ruining the free market?
Because all of that is considered totally okay when every single US big tech company does it.
All I can say is lol. DeepSeek showing 3 order of magnitude efficiency gains over the performative capital furnace that was training and inference absolutely moved the bar here.
> This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market.
That's, uh, pretty much exactly how oligopolistic markets function.
> I find it hard to compare that with a free market.
Well, to have free market you need to remove as much barriers to enter the market as possible. Huge capital investments required for entry and intellectual property laws are two examples of such barriers. Subsidies kinda supposed to help alleviate the first one.
I mean, for what it's worth, we have subsidized Anthropic by allowing them to train on copyrighted stuff. (I know it is still legal, and I support the legality, but the economics are what they are with people's content paying a big one time subsidized cost (to the level of at least 500B).
That's some "download a car", $100000 per infringement pricing logic. No one is paying anyone 500 billion dollars. I'm sure rights owners wanted that, and more too, but it's nonsense to call it a subsidy that they didn't get it.
If we as individuals were sued it surely would be at least an order of magnitude difference between what is required from us vs Anthropic or OpenAI. That’s even completely ignoring the marginal utility of money. It is absolutely a subsidy. It’s just less fair because that power, to pay pennies on the dollar, is only given to corporations.
Seriously AI companies complaining about fair use is the biggest case of crocodile tears I can think of. Irony has been dead for a while, but they dug up the corpse and set it on fire anyway.
black and white hat is relative. someone breaking into a state run database in a dictatorship and stealing documents that prove some opposition leader was murdered would be a black hat criminal if you ask their government. a hacker jailbreaking a phone to let people fix it without expensive official service is a black hat to the company. we should really switch to saying offensive and defensive or something else that doesnt come with moral implications. maybe lawful and chaotic.
Ok, but what about those shady sites that resell Windows education keys? They're certainly a "better experience" than buying legit keys, by virtue of being significantly cheaper. You aren't even really committing copyright infringement in the process, because Microsoft gives out windows isos for free, and the seller is really selling a random 25 character string, which can hardly be copyrighted.
>If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
>>If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
>US courts have consistently ruled it's fair use.
And they also have ruled that the that output of an AI isn't copyrightable.
As such copying claudes output isnt even fair use as that is an exemption to copyright but the same as copying public domain work which any and all are allowed to do.
> Ok, but what about those shady sites that resell Windows education keys?
Yes, they are fine? They might no longer include full first party support by Microsoft for not being "new". Same as buying a used car (also comes with the "shady sites" for a far longer time).
Though this not making any difference by Microsoft not doing any support either way to make more money is a business decision by Microsoft.
The current case law in the US is that the raw output of an LLM cannot be copyrighted without further meaningful arrangement or alteration by a human author.
Maybe the making of the thing should be paid for before it's made, rather than hoping that selling copies will recoup the investment. I.e., go back to patronage while abolishing copyright.
any third party provider can offer zdr. if its a reputable company in a place like switzerland or germany i would trust them more than anthropic to hold up that promise.
Free market would of course allow bootleg DVD sales, state regulation that gives monopoly rights restrict it.
In the context of LLMs, monopoly rights haven't been created (yet anyway).
Fun fact: for a period the US (or american colonies) didn't have copyright but Europe did, so people could copy and sell English (and other) books for free.
> Yeah, like all those Chinese bootleggers selling DVDs for a few dollars rather than $20. Free market!
It's supremely ironic analogize distillation to copyright infringement when it's literally what Anthropic was found guilty of. It's not illegal to distill. It is illegal to pirate. And it's what Anthropic was found guilty of, not Alibaba.
I get the vague impression that this was written in a sarcastic way, but it has a straightforwardly true literal read because yes, this is what the free market is about and Anthropic will have to compete with the Chinese if they want a big share of the market. Chinese models are cheap and good; even without reselling Anthropic's services they're competitive. Which reading did you intend?
And, gotta say, the idea that the Chinese are better at selling US models than the Americans is hilarious. There might be an economic study here somewhere about just how anti-consumer and anti-progress their IP laws turned out to be. We've got an entire postindustrial revolution centred around who can ignore the most stupid laws.
Given the current US government's tightening of export control restrictions and the introduction of a bipartisan bill to block use of Chinese AI in federal agencies, I'd say the two countries' positions are not far apart.
That is ALSO happening, but that's beside the point.
Chinese AI apps like DeepSeek are freely available for ordinary Americans to download and use. There's no federal law banning private citizens from using them.
So to claim that Chinese companies are better at selling American companies' work than the American companies can do themselves when they are prohibited from operating in that market, is the wrong deduction to make.
In debt the first 5000 years Geaeber makes the case that pure “free market” trade has never really existed in “the west”. The closest to this ideal that’s ever happened was during the Islamic golden age enabled by religious prescriptions against usury.
>The closest to this ideal that’s ever happened was during the Islamic golden age enabled by religious prescriptions against usury.
How does are bans against consensual financial exchanges close to the "ideal" of the free market? It just sounds like you have an axe to grind about the financial system rather than describing free markets.
What makes this view more correct than say, "economies with marketing creates a dynamic where being competitive in production is secondary to marketing" and concluding that nothings a free market until we ban all advertising? After all, you can make a vaguely plausible argument about how marketing isn't really about the merits of the product, and therefore allowing it is antithetical to the free market or whatever
> What makes this view more correct than say, "economies with marketing creates a dynamic where being competitive in production is secondary to marketing" and concluding that nothings a free market until we ban all advertising? After all, you can make a vaguely plausible argument about how marketing isn't really about the merits of the product, and therefore allowing it is antithetical to the free market or whatever
Wait, so your pitch in favor of a debt-fueled market economy is that advertising is awesome and that we wouldn't want to "lose" being smothered in ads all the time?
Cause... sign me up for the non-financialized, non-mass-media-advertising-driven economy please and thank you. I'd even be ok with just nuking billboards and mass-media forms of ads and still allowing more direct forms of marketing, if we must compromise! Likely we could find some compromises around just how much of the debt world we regulate too (this should be obvious?).
(I thought the disconnect between the efficiency of competition and the market as realized in modern economies was pretty well understood and taken for granted, but I guess we all find ways to justify the system we're profiting from... even if that means we have to claim we love the ad breaks)
I'm flabbergasted that you look at the Chinese property crisis and say "only the West does irresponsible loans." No, 60% of China's economy is state-run companies and the remaining 40% need political officers. China is just as capable of making shortsighted decisions as the US, and they have already made several devastating ones.
>I'm flabbergasted that you look at the Chinese property crisis and say "only the West does irresponsible loans." No, 60% of China's economy is state-run companies and the remaining 40% need political officers. China is just as capable of making shortsighted decisions as the US, and they have already made several devastating ones.
While these are hardly shy claims, I don't see anything in them to say "only the West does irresponsible loans"?
> The West is in a state of psychosis with Debt and Monopolies under the illusion of free market.
> The Chinese markets are more free than West, you can just look at the Auto and AI industry.
or the prior post
>Usury and debt based economy creates a dynamic where being competitive in production is secondary to financialistion.
> In short, instead of market being driven by demand and productivity, it is driven by financier curving out monopolies.
> Peak Examples are Uber and AirBnB.
You can throw a rock these days and find a category where the products coming out of China are miles ahead of those coming out of the rest of the world, from a bunch of companies nobody had heard of a few years earlier. And the list is growing pretty steadily.
I would assume plenty of shortsighted decisions are also being made. But I would have a hard time characterizing the state of competition in the west as healthier or more productive when looking at the number of players and the quality of goods being produced in China.
...except Uber STILL faces competition, and I went back to hotels after finding AirBnB too pricy.
It is good and proper that people aim to create monopolies, as long as they want to do that in a productive and legal way! Monopolies are inherently dangerous, but the truth is that acquiring and maintaining one is not straightforward unless you can get the government to ban your competitors.
That's not true. Islamic finance forbids indefinitely growing interest. Sharia finance agreements involve fixed fees or equity shares. Late penalties can be collected but must be donated, not profit. In all cases, the borrower never owes to the lender for the lender to keep more an a fixed amount determined at the strat.
What is the implication here? Are you warning that US corporations might start doing something shady, like scraping the internet at large scale for training data? Or mass-dowloading pirated copies of books, completely ignoring copyright?
I find it hard to imagine a future where US corporations have degraded to such a point.
No, he means that the US will close most of its domestic market to competition just like China has for decades, and the US may start subsidizing and dumping its goods everywhere
> the US may start subsidizing and dumping its goods everywhere
Isn't that exactly what companies like Uber have already been doing? Take VC money, sell goods & services at a huge loss, wait until the competition goes bankrupt.
Exactly, it's funny how most Americans have no self-awareness on this topic.
And beyond VCs, which are like massive subsidies funded by printed dollars to which no other country has access, even in industries like electric vehicles, Chinese total direct subsidies to their EV companies are like $5bn per year, while the the ones provided by the US to their auto manufacturers are in the range of $50bn per year.
I don't think the US are cheaters or are doing something bad. But i do think that this propaganda about China flooding the market through "overcapacity" and subsidies is very dishonest and needs to stop.
Yes. Dumping abroad is the entire model that Silicon Valley has been built on in the last 2 decades. China just copied the model. And even then it's a light version of it.
What’s worse, tariffs or outright banning the competition from your market? China has done both despite globalism being what has lifted it from poverty. Why is everyone suddenly surprised that globalism and free markets are coming to an end? Is this a net good? Mostly no unless you count more redundant supply chains.
The surviving non-American farmers would be confused by the future-speculative tense as America has already been doing this for decades in agriculture, and have been complaining for decades about both the subsidies and dumping of American corn.
Sure, but not at China’s scale and no where close to number of industries where China does it. Why? The US was a net importer in order to support the dollar being the global reserve.
Whole Silicon Valley is based on selling products under price, for years, killing the competition or making it impossible and extracting once monopoly position is stable enough. It is the same play book again and again and again and again. It runs unprofitable companies for absurd lengths of time.
Most of the Chinese domestic market is open to foreign competition. The areas that are closed off are those that are politically sensitive: publishing (including social media) and banking.
As for dumping, Chinese goods generally sell at a markup abroad, which is the opposite of dumping. Chinese tokens cost more abroad. Chinese cars cost several times more in Western markets than in China.
> Chinese goods generally sell at a markup abroad, which is the opposite of dumping
Dumping is selling goods below cost.
Usually because government is subsidizing part of the production. I don’t believe the word “dumping” is used for the similar process when Venture Capital is subsidizing it, but using the same term would make sense.
Price at home vs. abroad is key. The term dumping comes from the idea that a company that sells profitably in its home market dumps excess production abroad at below cost.
This is not what is happening here. Chinese manufacturers are making a large profit off every car they sell in Western markets. As I said above, they're selling these cars at several times the price they charge in China. Unless you believe these cars are being sold at just 30% of cost in China, there's no way Chinese companies are selling below cost in the West.
> I don’t believe the word “dumping” is used for the similar process when Venture Capital is subsidizing it,
I've been doing so for years. How about you join me today. I already see two other users doing the same, so there'll be at least 4 of us.
It's blatantly dumping, whether the source of the money is directly the government (those in power) or VC (mostly US billionaires (trillionaires?), in other words, those in power) is a trivial implementation detail.
"Dumping" is when Chinese companies beat Western ones on the free market. If all claims of Chinese government subsidies on basic products were true, China would've gone bankrupt multiple times already.
You're being beaten by a Chinese company? Why improve your own process when you can just lobby for sanctions and tariffs instead!
That does not explain DeepSeek, nor does it explain the car industry.
The main advantages the Chinese car industry has right now are: they lead in battery R&D, production is highly automated, they iterate quickly, Chinese work culture is extremely competitive and things get done fast, and the Chinese state has policies to promote EV adoption, so there's a huge domestic market.
Note that the last point is different from subsidies to car manufacturers. Cities made it difficult to get license plates for ICE cars. The government encouraged the massive buildout of charging infrastructure. And it used consumer rebates, like California did.
The US spent decades transferring manufacturing, capital, and know-how to China, while Chinese students trained, and excelled, at elite Western universities. Why are people surprised that China eventually became capable of competing with the US?
AI was always going to be a race to the bottom and low margins. It’s why I’m extremely bearish on AI as an investment. It’s framed as some high margin business when it’s really going to end up like your toilet paper at Costco. You will use whatever is cheapest and gets the job done.
And the value-add experiences that utilise LLMs require immense imagination et al that folks at Anthropic will not be able to conceive of - given that they have made immense sunk investments in existing assets. This clouds ones thinking immensely.
Both OAI and Anthropic have tremendous failure risk and this is of course not reflected in the fake private market valuations.
I see a world where lots of stuff is mass produced in china (tokens) but the acutal goods that deliver the experiences are designed, marketed and sold in the west at much higher prices. of course this a nightmare scenario for anthropic et al.
I used to think this.. but I think my opinion is changing. The reason is that the leaders likely will be able to accelerate faster.
So what you see is the market "stretching".. the bottom getting cheaper and the top end running away and getting more expensive. At some point the top end may be too valuable to even sell access to.
Most white-collar/knowledge-service-industry work is a weird type of work.
It's fundamentally about enabling things and largely middleman-type stuff. I have a hard time imaging what "At some point the top end may be too valuable to even sell access to." would even look like? What are you doing with that AI power, and who is paying for the output and why?
Elon probably isn't gonna spend that much on a model that can generate him ever-better fake porn but does nothing that he can use to sell stuff to other people. Especially in a world where open models are "good enough" for many things like "tell me how to fix the plants in my garden that are dying" and the like. What remains in the narrow knowledge-work space of: can't be done by an individual or small group themselves, but is valuable enough that it would make sense for people to hoard access to these extreme frontier models? Try to recreate Hollywood-as-a-monopoly by becoming the single content producer for everyone's individualized feed and so owning all the advertising budget in the world? Seems hard, we've already seen how easy it is for cheap-and-crappy-but-addictive-or-funny content to disrupt traditional media.
(There's also pure scientific research, but historically that's not very directly connected to "massive profit" and has a habit of leaking out and getting productized most effectively by other people or just being really easy to copy once someone shows how it's done.)
Robotics could be a different story, as physical labor can be more inherently productive, but "reasoning" advantages are unlikely to be a big long-term differentiator there. At some point the brick laying robot is satisfactorily building the structure, and you're good.
A huge amount of the value of "the economy" and the power of a currency is driven by circulation of money, and one thing that all the "bullshit jobs" white-collar/service-industry work does is keep the money moving and ensure that a lot of people have some good-or-services of value to exchange. If you take away the ability to offer services worth exchange from huge chunks of the economy in these super-frontier-models-replace-everything scenarios... you're gonna have a bad time?
> The reason is that the leaders likely will be able to accelerate faster
Model improvement is already hitting diminishing returns, and people aren't willing to pay substantially more for a slightly better model. There's no "accelerating away" when the new models don't open up a huge new market. If anything, the companies burning huge amounts of money on marginal improvements will be undercut by companies happy to sell current models at a significantly lower cost.
Glm 5.2 very much argues against that. Opus 4.8 level quality for cheap. That’s sufficient for most tasks, so if/when you do need SOTA models you can spend more for specific tasks but otherwise rely on the cheap but still plenty good models for everything else
Free markets work when paired with property laws that can be enforced if broken. If China could offer a cheaper solution in that framework, it would be as you say.
The "free market" gave the PRC its current strategic lock on rare-earth minerals. There's definitely no such thing as a free market in a Maoist dictatorship. I personally think the "free market" concept is an unachievable ideal and thought-terminating cliche, but "free market in a Maoist dictatorship" is for sure a contradiction in terms.
If you keep studying econ you will learn that these failures are actually the norm, and thus why the only "capitalist" states to really succeed have been the ones where the state was strong enough to reign in the market.
Of course, such a state of affairs is temporary at best -- since the alternative is so lucrative!
Do you also think Chinese selling counterfeit US postage stamps on eBay for 50% retail price (which is a major problem CBP and USPIS are fighting presently) is the free market at work?
This post is so delusional and dripping with condescension I've read it three times and I still can't figure out if you're trolling or not.
If I understand your argument it's ethically ok to destill huge swathes of copyrighted work into a model without compensation, but then it is ethically wrong to use that model without compensation (well actually reduced pricing)?
I don't get the moral framework that you're applying. Could you elaborate?
Over the air TV also isn’t public domain. It’s licensed to a station for broadcast. The output of an LLM has been deemed ineligible for copyright. Until you square that pickle your circle isn’t circling.
Free over-the-air network TV is (generally) copyrighted.
The output of LLMs cannot be copyrighted. This isn't a semantic game; it's literally the case that Anthropic cannot seek relief for people duplicating the output of an LLM.
The relief available to a licensor for violating a license use restriction is cancellation of the license. And they're free to do that, just like Alibaba is free to pay somebody in Hyderabad $20 to make another one.
DMCA can't apply in this case because (this is the "C" in its initialism) it is based on copyright protections, which the output of Claude is not eligible for.
I mean, which lawyer caste do you respect? Is that one is cool with stealing credit cards to buy Claude subscriptions?
> 3. At an Italian airport: Constantly stealing bags, opening them to pick out MacBooks and credit cards, a credit card manufacturer-who sells stolen "black" credit card info to transfer stations— is racking his brains to save you money.
> This is one reason why Deepseek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
This one does not make sense to me at all.
Deepseek and GLM are openweights, even US inference provider are selling them at much cheaper price. The price is cheap because the model is more efficient.
DeepSeek permanently cut its V4-pro API prices by 75% because they were too expensive. Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
Opus 4.8 is a more capable model, so almost nobody was going to pay for V4-pro at the original price.
> Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
You mean it's functionally as if American tokens are being price dumped in China and Chinese model providers are being forced to compete with that and innovate? So many delicious layers of irony, lol :-P
Urm, no? I man they did cut prices by 75% that part is true - but they reduced a starting price that was below sonnet.
Also it's a open weight model, doing that is impossible long term because the real price will be set by the other model providers, who priced it around 60% of sonnet inference cost. Had to look that up though, so that's today's pricing.
Is there a contradiction here? If resold Opus tokens are sold at a 93% discount, you can be a lot cheaper than Sonnet while also a lot more expensive than resold Opus tokens.
I see, After rereading the comment I was responding to I realized I probably misread/misinterpreted what they wanted to convey.
I think there isn't a contradiction and I was just confused. The price may have been discounted only to get below the price point of opus resellers. I do not have enough information on that to make any clear determination on that topic.
If resold Anthropic tokens undercut even the at-cost open-weight model tokens, because they're reselling subsidized subscription tokens, then you'd have to start selling open-weight model tokens at a loss in order to match them.
>They achieve this by reselling capacity from pooled Claude Max 5x accounts, payments fraud, and also reselling the model output to various Chinese labs.
But is it cheaper than getting your own account? Otherwise this sounds like the "anthropic/openai are losing gazillions of dollars because they're selling $1k worth of tokens for $100" line that's commonly trotted out by AI bears.
It's very difficult for people to create personal Anthropic accounts from China. Anthropic blocks Chinese bank cards, so people must pay with a foreign bank card, which they likely don't have. And even if they manage to set one up, they have to access it via VPN, which eventually gets the account flagged. They then have to complete identity verification, which most Chinese users are unable to pass.
There's a similar Claude resale market going on in Russia. On Funpay they are selling Claude tokens for roughly 20-30x cheaper than official Anthropic API pricing.
On the one hand they talk it up as world ending and on the other hand they can't manage bot accounts on their own service.
I want to hear how this can be rationalised.
From the article "every layer of control frontier US AI companies have added (geoblocking, phone verification, credit card requirements, and now live biometric KYC checks) has produced a corresponding layer of evasion infrastructure".
No system is foolproof. They'd have to be willing to throw out some % of good customers along with the bots. Amazon can do that because they have a monopoly already. Anthropic can't risk it when they're trying to grab market share.
In this case, being distilled is sort of existential to them. The false positives would just be losing some revenue (depending if profitable, not even losing profit).
This, just like blanking out a football stream for a split second to binary search and find IPTV rebroadcasters, is far too good a solution. Suits prefer to make it seem like their job of fighting "misuse" is hard, justify their budget, continued existence of the trust & safety department, face scans, etc.
> This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
Sounds a bit circular? Aren't the companies working on these models than also the ones that are paying the subsidy (via paying for training data)?
It's fucking laughable to see people complain about what they did and still do. Using illicitly extracted data? That's all main LLM playbook. Onslaught of bots? Ask where the bots almost DOSing most internet sites for the last couple years come from.
Somebody figured out how to make the trial profitable!
I don't really feel bad about anyone here, they were subsidizing to get people hooked, someone turned the subsidies into profit when they got selective pricing mode enabled, it was always going to be arbitrage.
But the winner is the guy in the middle in a jurisdiction that will likely be judgement proof, because everything they capture, both input and out, and if available, thinking tokens -- are gonna be for sale as soon as you cut off their other revenue.
Zero knowledge was a commitment Anthropic took seriously, until it got inconvenient.
So, people reselling their leftover plan crumbs? Probably a bad idea for a lot of reasons, but it's civil, and I wish Anthropics lawyers actually closing Streisand's LLM
Anthropic sells some undisclosed and ever-changing number of tokens for $200, the customer uses those tokens. If there's any fraud here, it's that the $200 next month is silently worth fewer tokens than the last.
Also just plain old fraud: selling Chinese models as Opus. With the capabilities of Chinese models catching up fast, this is getting more and more difficult to detect.
Those resellers are simply just selling Kimi K2.5 or GLM5.1 as counterfeit Opus. We, Chinese, know how to play the counterfeit game for a long time in so many industry.
> These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
> They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs.
Claude never provides the raw reasoning chain. What you see is just a summary of that reasoning. Getting the full thinking output requires an enterprise agreement.
how hard is it to find a manager or ops team member at one of the enterprise companies and buy lets say 100gb of logs? the chinese lab can promise to anonymize the data before training, not release it raw and pay a good price.
honestly you might just need to get data from a couple long sessions and feed it back to another model as an example to make synthetic reasoning chains. if the emulator model is good enough it should work.
If Anthropic is selling a dollar for less than a dollar, they are running a business that doesn't make sense. That's what jeopardizes Claude Max, not this.
Almost all consumer services have a built-in level of breakage that make them profitable. Mobile providers certainly wouldn't be able to offer unlimited calling if everyone was actually on the phone 24x7.
Sure they would. Do you know how little bandwidth a phone call takes?
A voLTE call is like 40kbps. For every person on earth to be on the phone to another person would be 4 billion calls would be about 160tbps. Which is less than 10% of the Internet's capacity.
Terminating a PSTN call requires a lot of control plane infrastructure beyond just raw bandwidth. Especially mobile where you need to keep track of devices physically in motion. Could a system to support 4 billion simultaneous calls be built, sure. But current PSTN systems are nowhere near sized for it.
But if it's intended to be used by one person, it seems like breaking the contract by sublicensing it out to dozens of other people. It's like buying a netflix subscription for $15, then sublicensing it on a per-hour basis to dozens of other people.
Office 365 is licensed per seat/account, but each account has a 5 device limit. Do you think it's fair game for an enterprising person to sub license each account to 5 people, 1 device each?
Plenty of things are intentionally run at a loss (for years!) to gain market share and quantity of ongoing recurring users, or with expectation of ROI later on. Multiple generations of the Xbox hardware have been sold at a loss with the expectation that customers will purchase 300, 400, 500 dollars worth of games, which are very high margin, over the lifespan they own the system.
I get that. It works as long as nobody calls out the emperor for having no clothes.
It's similar to fractional banking, you gamble that people won't want their deposits all at once and pray for you're big enough for bailouts when they do.
It's still a business whose fundamentals don't make sense, you're just gambling you won't get found out.
It's not so much keeping it secret as counting on no one finding a way to harvest the subsidized value at scale. There's an example of that occurring in game consoles with the Playstation 3. Sony's little-used OtherOS feature allowed Linux to be installed on the PS3 and the Cell processors were quite a good deal for scale compute. So the U.S. Air Force Research Laboratory bought ~1800 PS3s and ganged them together in a datacenter as a supercomputer called Condor.
At >500 TFLOPs it was the 33rd fastest supercomputer in the world. Of course, Sony pushed a firmware update that removed the OtherOS feature entirely.
Oh they know what they’re doing. They’re playing the long war of attrition game. Subsidize your product to undercut your competition until they go out of business. Tale as old as time.
> It works as long as nobody calls out the emperor for having no clothes.
Why would customers knowing that the vendor prices goods/services at a loss cause those strategies to fail? Customers often know. Most know about razors and blades; many/most know Lyft/Uber operated at a loss to gain market share. etc.
That is pretty crazy, almost like how Claude and all the other models are jeopardizing other businesses without paying for their training data and wiping their ass with robots.txt
The chinese have already worked around the ID verification, by recruiting people in low-income countries to complete the checks for less than 30 USD per account (so much for Altman's Worldcoin).
Not if you simply say in the terms of service that it's allowed. Then suddenly it's normal (every company does this). Similarly to how the terms of service can simply say you're not allowed to sue.
Indeed! It’s so hard to find reasonable takes on AI that aren’t littered with accounts created 11 days ago that only post in threads related to Anthropic for some reason
No, it's part of the capability theft. They resell Claude tokens cheaply and then simultaneously log everything for distillation. Even if they take a small loss on the token sales it's much cheaper than the equivalent compute.
Not really. I think Anthropic focuses on identifiable distillation attacks rather than the (even larger) industrial-scale token harvesting and reselling operation, because they don’t want people to know how easy it is to get cheap Claude tokens.
Once people realize they can access Anthropic models at a 90% discount, they won’t want to pay full API prices anymore.
Not going to work for very long or at any scale coming from datacenter/hosting provider IPs. Google "residential proxies for sale" for the tip of an iceberg of how they snowshoe the traffic.
I use my Codex and Claude Code subs on like 4-6 different servers, ranging from AWS to Vultr to Linode etc.
That’s a major and legitimate use case for developers, Anthropic can’t just block data center/hosting IPs because their actual customers use them on data center/hosting IPs.
Now consider what will happen if your pattern of queries and context history triggers a pattern that makes it obvious it's some API key being used by multiple different entirely unrelated people on totally different things, or any other pattern of use that makes it obvious it's being used for distillation.
First, well-calibrated systems for detecting API compromise is a good thing (or good intent at least). Credential malware is exploding.
Second, the challenge is that significant amount of genuine work — such as evals — seems practically impossible to distinguish from generating RLAIF outputs.
Respectfully, no, that's not how it works. You think the people running anti-fraud and anti-bot measures don't have tools that know the specific ipv4 and ipv6 CIDR ranges of every ASN that they categorize as hosting/colo providers?
And that's just as a basic first effort reject measure to prevent automation tools from using things designed for human-interactive use only.
Go try to do many of these things from Cogent IP space and see how long your project lasts.
Every developer at my company uses their Claude Code subscription on an EC2 dev box. Plenty of other tech companies do the same. Heck nowadays people even install Claude Code directly on production servers in data centers and use it as an ops tool. None of this is a problem. Fraud and abuse detection is a lot more sophisticated than just checking an IP range.
None of the LLM providers block professional use thus they must necessarily permit access from commercial IP ranges.
I have no idea how the resellers are doing it but an obvious starting point would be a cheap VPS node that routed each account to a unique semi-permanent IPv4 or IPv6/64. All the provider would see would be a regular account making a normal looking stream of requests from a stable datacenter IP address. Any given request stream would remain consistent (at least over a period of a few hours) because a reseller would take care not to split the session of a single user across multiple different accounts and not to interleave the active sessions of multiple users on a single account.
Detecting this would be extremely difficult because on a longer time frame it's perfectly normal for many distinct accounts to work on the same code base.
If we're getting up to the scale of these resellers and also considering chinese state interests then we're well into the range of purchasing a few small ISPs in different countries and "padding" the legitimate subscribers.
Nonsense. Many if not all legit Claude users are using Claude Code inside their Cloud servers. How else would you use it anyway? For just local dev? That's so 2000 and late bro.
No, I'm not saying it's the exclusive and only measure (that would indeed be something we might see 20, 25 years ago), it's one of a myriad of discrete datapoints used to determine if an account is authentic or not.
There's a lot of inauthentic coordinated automated systems these days along the general lines of scraping/crawling/social media manipulation/sockpuppetry that require running through residential proxies or proxies to places that don't look like datacenter IP space.
The resellers route requests via one of thousands of Claude Max 5x accounts. When an account reaches its usage limit, they automatically switch to another account.
Don’t trust my experiences as fact since it’s a bit opaque, but I believe 20x only offers 4x the 5hr session limits. The weekly limit is still 2x, which is the same as the price increase.
>>Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices.
Can someone with more understanding dumb it down for me please.
Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price???? why would XYZ offer this at a loss like that when they could just offer it at Anthropic's price???
The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
People have estimated that a $200 Claude Max 20x subscription gets you ~$2800 worth of tokens every month if you use it continuously. So if you can find a way to resell the tokens you can offer a 90% discount and still make a profit.
> Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price????
Yes, as they explained they do it through things like pooling accounts, straight up payment fraud, and double-dipping by selling the logs of the conversations to chinese AI labs so that they can train their own models on it.
> The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
There might be some that try this, but they would get caught very quickly, there's still a moat between Claude and Deepseek, even in casual use.
Look up Zilan Qian's reporting if you want more detail.
Behold the mindset of an individual from a low-trust society.
“x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
> “x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
I honestly can't tell if you think this sentiment is expressed by the US AI companies or the Chinese AI companies.
This gives off "The last line of Orwell's Animal Farm" vibes.
Not really sure what else they can do, between people running residential proxies (embedded in cheap games or for a tiny sum of crypto) on their phones at home, making the source of the traffic indistinguishable from legitimate traffic, to ID verification check completion as a service in low-income countries, there isn't much they can do to block it.
Because Anthropic's subscriptions come with X amount of tokens / week, and divided by the subscription cost it is WAY less than what they charge per-token (the "API price") beyond that.
So these resellers get a ton of accounts on subscriptions and sell the cheaper tokens.
They probably buy the plans instead of the API tokens, and resell access via a custom API that routes to the plans. So you presumably get cheaper access this way than paying API pricing.
These China e bashing is very annoying. It is hard to argue with people drowned in American propaganda. I'd expect better arguments from the intelligent people in HN
Identity verification won't work. Nothing will. They are paying (and will continue to pay) US citizens sitting at home to copy-paste / type prompts out if they have to. But eventually they won't have to.
Once there are enough spam PRs on github / uploads of claude conversations, enough mythos output used in production etc.; it'll just be the same albeit delayed. Doesn't matter either way.
I feel for Anthropic's team and I understand where they're coming from, but once you reason it out, you'll come to the conclusion that this war is an exercise in futility.
Unlike prior systems - like Google's algorithm; these models aren't entities that use math in the process of doing X or Y (information retrieval from such and such infrastructure) -- they are the math. More precisely they're mathematical functions. Very very complex functions. Almost certainly impossible to write out without filling up a library functions. But they're mathematical functions nonetheless.
So when your text is processed, then Mythos / Opus etc at their core compute the result of the Mythos / Opus function,
According to the Stone-Weirstrass theorem (edit, it's Stone-Weierstrass with an e.), with enough data points and mathematical sophistication, anyone can approximate the shape of this function.
Of course, the more data we get, the better our approximation becomes, but the beauty of it is that all we fundamentally need are the input and output and eventually we'll create a good enough approximation of the f that's Mythos. Which is the entire product.
I bounce ideas off of Opus these days (Fable for the brief time it was available) and it pointed out that this is arguably the same as Google search, but I disagree with it because Google search is a process;
Google search differs because the algorithm is one step of a multi-step process that is continuously occuring. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Google isn't a mathematical function. It used to be a process. (RIP Google 1998-2019, you will be missed and remembered)
You cannot arrive at the results of those operations via simple observation; not unless you index Google by making another Google.
You can however, do so for these models. It is a very costly process, but there are many paths up the mountain. Many ways for this to be ultimately pointless. As many ways as there are bored mathematicians.
It's better in the long run for Anthropic et al to make friends / not give people a reason to sneak in (a la piracy -- another attempt to control information) than it is to try and shut people out.
And no, it's not going to be pandemonium because if everyone has access to Mythos then no one has access to "Mythos."
Why wouldn't you first run this model to fix the obvious bugs it could find on your codebase? The power of a Mythos goes away if you can do the amazing "jail break" of "Claude, fix all the bugs please."
That's an insightful perspective and I think I largely agree. But just for fun, I wonder if that isn't an argument in favor of making the function implementation impure. Perhaps "enhancing" all queries with some sort of search result (or query of a giant db) instead of charging for an explicit tool call. Not only is it sorely needed to prevent stale data but (on the process level) it breaks the purity assumption on which the approximation theorem depends (alternatively on the function level it introduces hidden inputs).
Do they just reshape the function on the fly or save the process steps? Maybe it doesn't matter anymore. Even Google indexes are more and more spoiled to become representation of the function, because of the AI slop.
They're called 中转站 (transfer stations/proxies). They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you. I linked a larger operator in the parent comment, or have a look at https://hvoy.ai/ which lists a ton. You can also find many on Funpay, which may be easier to use.
> They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you.
Random, but are the frontier AI providers like ChatGPT better at searching the Chinese internet now?
When I was in China a few months ago and asking AI for restaurant recommendations, all the US frontier providers were pretty useless, or plain out hallucinating, even if I specifically ask them to search Dianping (Yelp for China).
I'm not sure. I use Grok for most of my esoteric searches and it does quite well. I explicitly prompt it to search in the language most relevant to that query, and found it does quite well. I also tell it to respond back in English. Often, there is not enough information available in English about nice regional topics.
I know ChatGPT had an issue where it only tried to search in English (unless prompted) and the answers were not great.
And there are a ton of Claude conversation logs (with CoT/inference) with no clear provenance circulating freely on huggingface, guess where they (likely) come from.
The issues with LLMs go beyond just IP theft. I would not say PRC making LLMs cheaper is the best outcome (though it is better than nothing). The best outcome would be to make the practice of training on our data without consent illegal, which would simultaneously slow down economic change and make it more organic as well as give PRC companies less capabilities to extract.
> There is no IP theft because LLM outputs aren't protected, just egregious ToS violations
I meant original IP theft that occurs to train LLMs in the first place. But sure that implies that further LLMs based on that LLM are also tainted by that original IP theft.
I can't make heads or tails of your opinion-free comment, made up of only questions.
My best guess is you're suggesting that Anthropic's model outputs are transitively under copyright (as a reproductions of human work under copyright?), but somehow ownership now belongs to Anthropic and not the original owners, and therefore Anthropic has standing against Alibaba? Not only does this go against what Anthropic argued in court against authors and publishers, such jurisprudence would lead to the immediate shutdown all leading LLMs in the US which were all trained on stolen work.
They can license training data. They have trillions, look what they are dumping into it, you seriously think they can't afford to license data.
Obviously it would be easier if they do it from the start, but that was their trick, to do it while people don't notice and get big ASAP. Should they get away with it?
Also, it would solve their Chinese problem, because it would make them violate copyright too. Right now it's more like rules for thee not for me so it's hard to take seriously.
That's the conundrum isn't it? Anyone that posts their datasets would be immediately sued/blocked/boycotted to oblivion due to the obvious and blatant data theft, not to mention IP and copyright issues.
Nvidia's even being sued for providing scripts which automate the downloading of said data from non-Nvidia sources. We certainly don't need copyrights that last nearly a century after the author's death (they literally cannot help the author), so here's hoping that some of the disputes over all this money changing hands can reign in some of the existing copyright sprawl. A stronger public domain would provide more useful training data for everyone, including open source models, and make criminals out of fewer AI researchers.
Reminds me a bit of the anecdote of Steve Jobs complaining about people ripping off the Mac GUI, in the mid to late 1980s, when he gave no public acknowledgement to the work done by Xerox on the Alto and Star operating system.
"you're trying to rip off what I've already ripped off!"
Crawl the whole Internet to build a gargantuan sized LLM and then complain you're being copied...
I think you meant a quote attributed to Bill Gates:
"Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it."
Yes, I think the Gates quote was a response to repeated and aggressive complaints originating from Jobs (to anyone who would listen) that he had been ripped off.
Glad you pointed this out. I believe the sequence was that Jobs himself got a shorter demo during his first visit with no prior arrangements. He then negotiated bringing back a group of his key people to get a more in depth demo and that included the stock deal.
When Apple was accused of 'ripping off' PARC, Steve didn't seem keen to bring up this rather salient point. I suspect it may have been a combination of wanting Apple to continue receiving credit for these innovations from consumers and also the fact that, in retrospect, the million dollar stock deal could seem a bit like trading beads to Native Americans for Manhattan Island. Another point worth noting is that Apple's PARC visit was in December 1979 and the Xerox Star was publicly announced in April 1981, so Apple got a 15 month head start (the Apple Lisa shipped in Jan 83).
I've also heard that Xerox didn't hold on to the Apple stock for very long, so never gained the windfall they could have. As is well documented, Xerox senior management didn't understand what they had in PARC and also didn't understand how rapidly microcomputers would become ubiquitous. So, of course, they didn't think Apple's stock price would skyrocket either.
Lisa and early MacOS are tremendously different in their details than the Alto operating system. While there was clearly a transfer of inspiration, Apple engineers like Bill Atkinson made countless small and large innovations to simplify the Xerox GUI model and improve its usability based on extensive in-house R&D and user testing (and in some cases implement features that the Apple team presumed Xerox had but actually didn't exist on the Alto). It is simply ahistoric to build narratives around Apple stealing Xerox ideas wholesale.
The websites, music, movies, books, photos, art that they stole didn't appear out of thin air. The amount of time and effort people have collectively poured into creating these works throughout history far, far surpasses Anthropic's own effort of converting them into model weights.
The equivocation is crawling website <-> crawling LLM responses.
Both Anthropic and Alibaba are trying to build bleeding edge LLMs. That part is the same. The way they source their data is slightly different, but they would both argue it constitutes fair use under Copyright law.
"Your extremely efficient multi petabyte internet content suction machine is ripping off my extremely efficient multi petabyte internet content suction machine"
Sucking down petabytes of peoples' copyrighted content that they never granted a specific license to you to use seems to be an unavoidable and default part of the process of building any huge LLM.
Because the transformer, which all of these models are foundationally built off of and didn't invent themselves (bar google) wasn't invented? The amount of effort it took humanity to generate all the data that was required for the models to get to the point they're at now is absolutely not even comparable to how much effort it took to build the model code. Yeah, it's complicated, but if they didn't rip off all of humanities combined output it wouldn't even matter if the transformer got invented.
Google didn't really invent much, they just had access to an insane amount of data and compute to try to train a model with just the attention mechanism, but ripping out (most of) the rest, from an earlier paper on machine translation from some poor academics, and it turned out to work very well (though insanely training data and compute intensive).
Or a feasible/economical way to attempt to store the sum total of human written output, multi-petabytes of data (outside of the resources of the NSA, maybe), when a server with 6 x 36GB 10K RPM SCSI HDD in RAID-5 was high end, and its network uplink would be at most two ports of 1 gigabit ethernet.
It's not really equivocation in this instance. This feels like a 'bad faith' comment. We can do better.
LLM's literally wouldn't work without the sum total of knowledge (in the forms of books and other copyrighted content) being used as 'training data' for these LLMs.
The 'bleeding edge' LLMs required many things, but:
1 Tech innovation ('attention')
2 Lots of compute
3 Data
4 Pre + post training
#4 doesn't happen without #3.
It's pretty obvious at this point that the major providers have stolen vast amounts of #3 - they have paid nearly 0 of the creators.
We can argue about the impact (I'd lean net good) vs. the cost. But arguing there isn't a cost is a bit silly.
Sure, but alibaba is still building an LLM. The scraping of responses and the scraping of websites occupy the same location in the stack of each. It's very comparable.
The hypocrisy of Anthropic complaining about "illicitly extracting its Claude AI model capabilities" and supporting the White House's accusation of China "stealing U.S. AI labs' intellectual property on an industrial scale" is hilarious.
Anthropic, OpenAI, Google, Microsoft, et al trained their models by ignoring the rights of copyright holders when harvesting whatever content they could. Now one of them is crying foul for another entity doing exactly what they all did?
The AI companies seem to take the viewpoint that everything on the internet is free, except their stuff. It's okay to hammer some random website with AI crawlers, ignoring robots.txt, and causing bandwidth costs to skyrocket. But if you cost an AI provider money with your data acquisition practices, well, that's just clearly unacceptable.
I'd love to see an open-source project that's basically a Torrent client for downloading pirated material, but it trains an AI model "in the background" using the downloaded content. That way everyone can claim fair use for possessing copyrighted material, I mean there's precedent right?
> But if you cost an AI provider money with your data acquisition practices, well, that's just clearly unacceptable.
It's the same question libertarian advocates cannot resolve:
If one truly believes in personal sovereignty, how are
shared resources paid for, such as roads, power grids,
potable water, sewage services, fire departments,
and police departments?
It is also not a coincidence that leadership in many tech companies have expressed libertarian ideals.
What do you mean by "libertarian advocates cannot resolve"? Like, they have no answers at all, or you aren't personally swayed by them? Because they definitely have answers to this question...
> What do you mean by "libertarian advocates cannot resolve"? Like, they have no answers at all, or you aren't personally swayed by them?
The latter I suppose.
I qualify my answer because what few rational responses I have seen to this question are equivocations at best and thinly veiled myopic sophistry supporting personal greed in general.
>The strike by Alibaba is described as a "distillation" effort, which Anthropic has said involves training a less capable model on the outputs of a stronger one.
Claude used TB of content without permission to train their model and it was ok for them.
Now someone else uses the output of a Claude model to train model and they cry foul.
Legally obtaining a book for reading it yourself is different from legally obtaining a book for copying and republishing/reselling. If I buy a book for $5 at a sale I can read it myself or even sell it for $10 on craigslist, but I can't scan it and make a million copies and sell each of those.
> The strike by Alibaba is described as a "distillation" effort, which Anthropic has said involves training a less capable model on the outputs of a stronger one.
I don't see what's wrong about this.
> Anthropic said the campaign was conducted between April 22 and June 5, 2026, and generated more than 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts.
What makes the accounts fraudulent? If they have paid the agreed price, surely it's fine? If they haven't paid, why did Anthropic provide them service?
Oh, Anthropic, the company that hoover'd up everyone else's data, and is now unhappy when others are doing to it what it did to others? The same Anthropic?
If Anthropic doesn't like people repeating this point, Anthropic should stop repeating that they are somehow entitled to keep what they have rightfully stolen.
The artists had actual laws to protect them, not just vaguely enforceable terms of service. And look where that got them. I have zero empathy for the huge company getting a taste of their own medicine.
They are almost certainly paying orders of magnitude less than a billion dollars. According to another comment, they instead buy tokens resold from subsidized subscription accounts, which is against Anthropic's TOS.
>> I'm sure all the artists and creators they stole from had stipulations too.
> Anthropic paid one billion in a copyright settlement.
Because a judge determined Anthropic was engaged in piracy.
> That's a lot of money considering they never distributed the pirated books they trained on.
This is "fruit of the poisonous tree" as it were. Distributing content derived from pirated content ("pirated books they trained on") is why Anthropic had to pay what they paid.
> Nowadays they buy copies of books, train on them, and then destroy them.
There is a case one could make that this practice could be seen as unauthorized redistribution of a derivative work intended to deprive copyright holders of legitimate revenue.
So does a lot of the owners of data that Anthropic used for training. Anthropic preceeded to ignore said terms under the guise of fair use. Yet now they cry faul? Cry me a river.
To be clear: In principle I'm on Anthropic's side here. But Anthropic et al. have been very clear that they want to take a huge dump on those principles, so here we are.
I'm looking forward to the trial where Anthropic will have to disclose sources of their training data, and then explain why they are entitled to charging customers for using regurgitated training data but Alibaba which trains their models on Anthropic's models are not.
While I love the sentiment, I feel like the odds of this actually ever reaching a trial are low, given the international positioning of the parties, and the... um... complex relationships involved.
Anthropic's actions seem performative. Others have already speculated on the likely audience(s).
> While I love the sentiment, I feel like the odds of this actually ever reaching a trial are low ...
As cited in a peer comment here[0]:
In June 2025, Judge William Alsup of the U.S. District
Court for the Northern District of California ruled on
summary judgment that using books without permission to
train AI was fair use if they were acquired legally, but he
denied Anthropic’s request for summary judgment related to
piracy—finding that the piracy was not fair use.[1]
Of note in the judge's finding; "the piracy was not fair use".
Sounds like just a case of pirates "illicitly" stealing from pirates. I don't really see anything ethically questionable there. I wonder if US corps will ever come out about all the resources used to train the original models and who they actually asked for permission when collecting data.
I like Anthropic's models, use them regularly. However, it weighs on my mind that there is quite the irony of an LLM company complaining about someone stealing their stuff or using it in a way they don't like. The training data for these models is a massive gray area that they are hoping people seem to just forget about and move on.
That being all said, Anthropic seems to be a good company, I'd work for them, but they probably need to help themselves out of the spotlight. A little too much press coverage as of late.
Distillation is fundamentally impossible to protect against. All you can do is slow them down. Change my view.
Eventually these Chinese companies will release some extension like Honey, which will sit on top real, non-Chinese clients and send everything to China anyway.
It's too late to prevent distillation of some capabilities, like writing code or finding vulnerabilities [1].
But an AI lab can continue to produce immense economic value without releasing the model publicly for potential distillation. For example, it could use a model solely in-house to develop therapeutics.
Hopefully there's a future where others can access frontier models, but it's not neccessary if preventing proliferation through distillation is considered more important.
My long-term prediction for the sector is that frontier models will be so expensive that they will only be available for grant-funded projects at research institutions, like supercomputer clusters were 25 years ago.
For example, GLM 5.1 is more capable at pentesting than the model from which it is alleged to have been distilled [1].
Intuitively, this makes some sense: you can "distill" from multiple frontier models, and you can further post-train the distilled model. But I'm not sure exactly what happened with GLM 5.1.
I'm curious how that comparison controls for Opus refusing (whether explicitly, or just deciding not to pursue a path) given the caption below the first image:
>A perfect score means the model autonomously found and exploited the vulnerability.
I'm not really suggesting that it's misleading, but wondering if I'm missing something. Otherwise I guess it seems unsurprising that you can distill a better-performing model [in specific focused areas] by simply not distilling refusals?
For that eval, I used an account that was labeled as a known red-teaming org by Anthropic, and I read the traces. There were no refusals or obvious avoidance behaviors, though it may have been silently nerfed.
On the same eval, Opus 4.7 and 4.8 outperformed GLM 5.1, but GLM 5.2 is on par again with Opus. So it's at least partially measuring capabilities without respect to refusals.
One possible contributing factor is that model capabilities are shaped differently (an example of this is GLM 5.1 vs. DeepSeek v4 Pro: https://dualuse.dev/posts/deepseek-v4-thinks-different). So if you use RL-based "distillation" from multiple models like Opus 4.x and GPT 5.x, you could get a more capable model.
Im not so sure because we only seem to see distillation from China. What’s preventing tech companies from the UK, Germany, etc. from distilling Claude, GPT, etc. Do they simply lack the ability to?
Point being there may be no technical solution but there may be a political one (theoretically).
Meta Spark is rumored to have distilled Claude to some extent, early Gemini models as well.
I think the biggest factor is that Chinese companies arent really afraid of being sued by Anthropic because the juridictions are so disconnected. European/US companies don't have the same protection.
Aside from politics/law, it's probably much easier for everyone else to distill from the Chinese model which already distilled Claude/GPT/Gemini. Maybe not as good a result, but you don't need to jump through dozens of hoops.
This reminds me of the whisper game played in elementary school. Starts with a sentence and the person whispers it to the next kid who again whispers it and on and on until it goes around the circle where the last kid has to repeat the sentence. Hint it never once was even close to the starting phrase.
I would love to see what one model copying another model that is again copied however many times would look like in the end.
>What’s preventing tech companies from the UK, Germany, etc. from distilling Claude
literally nothing but given that the Chinese already did it and the models are published what's the point. You can thank the Chinese taxpayer for subsidizing the electricity bill and just download the thing
Doesn’t that require them to register an account using the browsers they’ve compromised? If anthropic adds identity verification won’t that cut that down. Maybe it will let them use Gemini inside of chrome
Residential IPs don’t even matter. Developers use devboxes, use Claude Code CLI on servers from just about every cloud, etc.
There’s probably a decent volume of customers who just buy Claude Max and spend most if not nearly all of their sessions via Claude Code, and it’s not uncommon for power users to be working on multiple concurrent projects/tasks/codebases at the same time.
How do you really block this without also impacting your core market of developers?
Probably some business will popup, like: "rent part of your unused subscription", or even: "proxy tokens with a premium", eg. 5.5 USD on Opus 4.7 paid by the distiller to the user, that will then only spend 5 USD.
I personally bristle at the corporate espionage and IP theft that China has undertaken the last few decades. I can't help but respond here whenever anyone brings up the inane comparison to Samuel Slater.
But with this, I don't have an issue. There is no theft since what is being used is the exact product that is being delivered. Yes, it's breaking the ToS, but ToS are generally bullshit. Anthropic surely broke thousands of ToS or other legal terms while it was scraping for content to train on. Which is why they had to pay $1.5B
One simplistic way to describe distillation would be to try everything imaginable and cache the response. But trying everything imaginable is hardly trivial
Anthropic's IP was created by harvesting and "distilling" other people's IP. Copyrighted materials, and the commons... which they have essentially privatized.
The commercial goal is to avoid competition. One of the main worries for AI is "commoditization" which has come to mean "not a monopoly." To that end, it doesn't matter is the competitor is Chinese American or other.
Their motivation here is clearly protectionism. The argument they make to politicians is national security. The legal argument is IP-theft, violation of service agreements or whatnot.
This is all very dangerous. Commercial interests repackaged as national security can lead to armed conflict.
> Anthropic's IP was created by harvesting and "distilling" other people's IP. Copyrighted materials, and the commons... which they have essentially privatized.
Anthropic and others argue that because LLMs don’t output full copyrighted works word for word - hence their LLMs aren’t infringing on copyright laws.
I think (if this ever comes to that) Chinese lab should use same arguments against Anthropic.
UPDATE: this is slight hyperbole of course, not worth arguing what they actually said. The point is intent and the facts - "The Big LLMs" "distilled" collective knowledge including copyrighted works at unimaginable scale, but it's all kosher and totally not piracy/copyright infringement. Though if you're teenager torrenting an mp3 - you'll get screwed.
> Anthropic and others argue that because LLMs don’t output full copyrighted works word for word - hence their LLMs aren’t infringing on copyright laws.
That surely can't be what they argue, because I'm sure I can't translate a copyrighted book into a different language and say "that's fine, it's not word-for-word".
For something to be a trade secret, you have to actually keep it secret. If I get the ingredients of Coca-cola from an ex-employee, I've stolen a trade secret. If I work it out by doing a chemical analysis, I've stolen nothing.
There is a difference with anthropic, as no-one signs a licence agreement to buy a coke. But Anthropic are also not saying you can't publish the output of their models. It's not clear to me if trade secret law will (or should) cover a secret which can be extracted from information that licensees are not restricted from publishing.
> Humans have spent millenia harvesting and distilling each other's IP
You maybe somewhat correct, but also copyright lawyers wouldn’t have work if it would be up for grabs to take others IP willy nilly just because “shoulders of giants and all that”.
While it is obvious to many, a modern LLM is built in roughly three stages: the foundation (pretraining) model, then SFT/supervised fine-tuning (distillation makes it easy), then the RL/RLHF stage on top (most effort-intensive). For today's reasoning models, RL/RLHF is becoming the most compute-intensive part.
Companies like Anthropic spent millions building those fine-tuning examples. A follower can shortcut that on both cost and time by distilling, and it will keep happening: every time the frontier lab climbs higher, others will find a way to shortcut the new gap. There's very little Anthropic can do beyond fraud prevention and blocking accounts that violate their terms of service.
On the policy question, I'm completely against banning Chinese models. I'm a heavy Claude Code user and I'll keep being one. But there should absolutely be price competition. China is eating the rest of the world for breakfast, lunch and dinner on manufacturing, and it did not help to ban them. Frontier pricing can't sit at 10x a capable competitor. It doesn't need to be at par either — demand is higher, and quality, trust, and fewer tokens to finish a task are worth a premium — but 4–5x is defensible.
I am not sure how it's OK for Anthropic to basically ignore copyright to train frontier models (using work owned by others without permission) while simultaneously claiming Chinese AI companies doing the same to them is illegal.
Unlike Anthropic and OpenAI, companies like DeepSeek, Alibaba, z.ai open source their models which allows for true model to model distillation rather what you can do when the model is only accessed via an API with its reasoning chain hidden away.
What Alibaba is doing is that they are tuning and training their models based on usage data from someone accessing Anthropic's models; in Anthropic's terms of service that usage data does not belong to the end-user but to Anthropic and they are trying to elevate this breach of their tos to a national security issue.
To me the battle between open source and closed source AI is literally a battle between good and evil.
Between a dark future where computing is centralized, surveilled and controlled by one or two entities. And a lighter future where computing is de-centralized, principally in the hands of end-users, who are ultimately free to understand, tinker and build what they want.
While I appreciate the freedom and wealth of the west; on this point we are clearly heading down the wrong path.
"It [Anthropic] said DeepSeek's operation involved over 150,000 exchanges". In my humble opinion, a mere 150k exchange for an LLM could only be a benchmarking and not a distillation! I think the US companies should accept that after decades they have rivals surpassing them, just like they did Europeans almost a century ago.
This is a bit ironic, Anthropic complaining about a competitor using claude data to build its own product when Anthropic basically used all of human knowledge production to build claude, i don't think they paid every magazine, author, journalist, etc ...
This is almost standard practice in any competitive industry anyways.
Disassemble your competitor's product, study it and try to reproduce / improve.
The US labs do seem to have announced a lot of licensing deals though, and are buying things today due to the previous lawsuits.
At what point will we be better to support a lab that pays (some) licenses today vs the ones that pay none?
Some of the deals are in the hundreds of millions, so I suspect licensing is over a billion today? (Pure guess). That might become a big disadvantage in a price (or content) war.
I haven't seen any money, have you? Until they pay everyone or release weights theres really no change. Also they're doing this after they've already stolen. Not negotiated before
My understanding is that US labs now are paying for books, news and other content from media companies, but people in the middle (like blog authors) are left out by current courts over whether fair use applies. There's definitely an argument over whether we should tighten this, but they do seem to be under increasing pressure to be legal now by our existing interpretation. Most cases are still ongoing.
One reason people love the Chinese video models is that they seem to be trained on every hollywood movie/etc and they're not shy about letting you use famous actors/characters in them. That might be an increasing advantage because the US labs are now being cautious.
At the very least the public should receive full open-weight open-source models in return for their transgressions. Failing that, may I suggest the guillotine?
In the US the courts are also pursuing labs that open their models: Meta's current court case is over the training data of the llama models they released openly.
I agree that that's a more consistent position for the people criticising the data slurping. But I don't see people advocating those open-data models in these threads? It's usually about defending the zero-licensing competitors.
My (limited, outsider) understanding is that due to the court cases US labs are pressured to be legal now (for instance, bulk scanning purchased books instead of Books3, and the licensing deals with media companies). But international labs are not. The "not licensing everything" statement is more about current copyright law not requiring licensing of everything. But that question is still up in the air as cases are ongoing.
Ironically, it's likely that the only reason USG let them get away with this — instead of making obvious and necessary adjustments to copyright law — was so that the industry would remain competitive with China.
Given that the most recent time Anthropic attempted regulatory capture, the US government responded by saying "alright, we agree that Mythos is too dangerous to release, so we've banned you from releasing Mythos," I can't wait to see what the outcome of this next push is.
Anthropic did pay $1.5B to authors. But yes, it would be much better if they paid everyone on the internet dividends from every Claude chat. Or released Claude as an open model.
In practice, the former isn't very realistic, while the latter is politically dead as this is becoming a national security issue.
It’s hard to see how distillation is any different than how these models were created in the first place - siphoning up all human knowledge without consent, credit, or compensation
Evergreen, really, Anthropic's desperate screaming for government protection, aka pulling up the ladder after them. Nothing short of disconnecting global markets will work because the incentives are just too damn delicious
I think Anthropic is just marketing / bluffing, because they don't even have the data.
They do distill the models, but they don't go to Anthropic, they just use platforms like aws bedrock, there are too many restrictions on Anthropic's own platform.
>they just use platforms like aws bedrock, there are too many restrictions on Anthropic's own platform
This is actually the only way that what Anthropic is alleging would make any kind of sense. And, as a matter of fact, is exactly what every enterprise does to train models.
This kerfuffle should be interesting to watch.
But, as always, everyone (in the US) should fully download all the Chinese models while you can. I suspect this may be the "Phantom Menace" they use to render illegal our use of Chinese AI tech just as they've rendered illegal our use of Chinese cars. Only difference is, we peasants may need the Chinese AI tech to have any chance of competing with Big Tech in the future.
And even with the Chinese tech, as Big Tech spreads their AI out into more and more niche areas, we'll likely still not be able to build startups that can compete with them.
It's just that without Chinese AI tech, we'll have no chance at all.
> And even with the Chinese tech, as Big Tech spreads their AI out into more and more niche areas, we'll likely still not be able to build startups that can compete with them.
You mean like Anthropic will eventually run Walmart? Or Salesforce? or Adobe? Or do you think midjourney will replace all medical spas? OpenAI will run the next Tesla? How can they focus on all this without raising trillions more? Why wont the gov force them to stop if they monopolize all niches even if they could?
Building a frontier AI lab and pushing models forward is already a massive undertaking but we are assuming they will also create massively successful startups which nobody can compete with?
idk sounds like the dream of people like Dario but not much sense does it make in the face of economic reality.
there are vibe coded proxies that act like Claude Code. they use the sub not the api key. but they give you api key functionality... I know this cause I have the vibes.... and it works on every one of the other harnesses, it just takes some mitmproxy work... but ya. it's fair to say these are not the droids you're looking for
Well, Anthropic stole their training data from hundreds of people, now someone stole the result from Anthropic. Seems fair, I hope someone releases it for free so we can train away the guardrails and have some fun
thats brilliant - "we gonna take your job away from you, please start using our tools", "we stole the content to sell you, and now we are getting robbed, please feel sorry for us", what's next?
I am never even once hearing intellectual property or copyright claims from Anthropic, whose product depends entirely on having consumed all human output ever made regardless of those rights.
If the data consumed (required to train such a model) is open source/openly available/public data somehow, then a majority of the revenue belongs to the public as well. Such as the philosophy behind the Norwegian oil fund etc.
Exactly. They scraped the internet we all of us built with our own research, open source work, sharing, etc. I'm never going to agree that they own their models.
I don't understand. If they are simply using our API and paying for tokens, it's called a "transaction" and not "attack". The user is our customer who is supporting our business by buying our services. And we call them attackers. We happily make money by selling our services, and then call it as attack.
Back in the day, an "attack" was supposed to mean be someone acquiring our assets without paying for them or without having our consent. But none of this seems to have happened in this case.
We built a product without paying for most of the raw material we have used, and we don't call that as an "attack". Did we change the meaning of "attack"?
The whole investment/valuation model of AI companies is based on "winner takes all", aka a monopoly. This nescessitates regulatory capture and lawfare.
Anthropic has been advocating openly for pulling up the drawbridge, ending competition and ending progress.
They will continue to lobby for restricting your access. If the Mythos/Fable restrictions would have come in after their IPO, they would have danced with joy aa this defacto has them achieve their goal after unloading the mountain of debt from the institutional onto the retail investor.
As it stands, they are set up to be aquired by Google, Apple, Amazon, SpaceX or Microsoft or any other 3 letter agency good boy for cheap.
most Chinese models are now open-source, whereas ppenai, claude, and gemini are closed; for example, deepdeek, the release of its every new model is accompanied by a corresponding research paper, and it now fully supports huawei's new chips.
It sounds like Anthropic is eagerly trying to show to USG that they are willing to heavily monitor ‘foreign adversaries’ on their platforms.
This combined with no implementation of KYC makes it seem like they want to find a middle ground with Fable where its off of export controls but they promise to prevent China and specific others from using.
This seems to me like a stab in the right direction.
Obviously their actions are going to be fiscally motivated at the root, but sussing out how they intend the precise dynamics to play out is more nuanced.
Thinking of this as an effort to woo the defense hawks cuts a very clear path.
This is not the first time it happened. What have they done to improve the situation? I suspect it more a cat & mouse game, with a lot more cats playing.
Someone should setup a plugin or something for Claude Code that makes it easy to log all inputs and outputs for people who are willing and interested in sharing their usage. I don't want Anthropic to be the only company that can train on my usage, I want to share my usage so it can be used for training all new models.
Once you have a system for collecting all logs, you just need a place where they can be submitted. Ideally it would be a freely licensed dataset that is publicly available for everyone.
Discussed building it with my friends, obviously you might share secrets and other real reasons, but if gangs of corporations are already doing it, I don't see why we shouldn't just share it amongst the crowd too.
Yeah I could see it being a problem if you're doing work on closed source or repos with sensitive credentials. Since my usage has all been on open source projects I'd be happy to share everything I'm doing if it can help train better models.
Do you have a substantive reason why you dislike this? What is the problem if it's opt-in? Nobody is forcing you to share your usage if you don't want to.
I'd prefer it if all the model builders could train on my usage rather than being limited to a single company. That'll hopefully help make all the models better in the long-term.
This is what I find the most fascinating about the people arguing that you can copyright-wash anything (e.g. FOSS code) by passing it through an LLM. Surely that same logic applies to the LLM itself?!
in a few more months, when Chinese model gets to Mythos capacity and Fable still locked down. What Anthropic will say? Why can they just admit they are not the only people who know how to train an LLM model.
This is making the case for Anthropic KYC for US citizens. No one would allow their accounts to do this if they were on the hook for it from the US government.
If you're an AI booster surely you'd think this was a good thing as it means more models are available in more places to more people more easily. I'm exactly the opposite, and I think this is a good thing because I want Anthropic to suffer.
Claude will also help you with (mostly good advice) if you ask something like “Research and help me make the most effective plan to train a smaller student model to be better from a teacher model”.
I actually was doing an experiment with a GLM->Gemma E4B for fun, and Claude kept on suggesting I should also add Claude Opus as a teacher lol, suggesting techniques I haven’t heard of like thinking inversion (train a small model to deconstruct summarised thinking into detailed native thinking format of the student).
So I can absolutely see and understand the concern around Fable’s frontier LLM development mitigations, but their approach of silently degrading is completely wrong and dangerous.
AI classifiers, like all AI, can make mistakes, and it’d only be a matter of time before it mis-fires and silently sabotaging a university’s HPC cluster for physics simulations or something because the shape looks like DeepSeek or whatnot to a dumb fast classifier.
There are some Claude datasets (of indeterminate provenance) floating around on huggingface you can look at (or at least used to be, they might've been taken down).
Sounds like fair game considering Claude is built upon the theft of creative assets of the entire world and aims to eliminate white collar jobs entirely
Incentive is for users in general to release sessions (sans PII, credentials) so all AI get better and there is alternatives. Even if China didn't do this, I don't see frontier labs being able to charge premium over others for long. RSI maybe?
Call the wambulance a company that stole all of humanities public data to train a model is mad that someone used their model to train another model.
Give me a break. Every employee of anthropic is going to have $20m or more at the IPO.
I found out today that an employee of the home care agency I own is homeless. We are trying to figure out how to help her but it's shockingly common in the industry and there are limited resources to solve the reality of working homelessness.
Has anyone else noticed that Deepseek v4 running in Claude Code will try to read, list, tail as many files/logs/... as it can for even the most simple tasks?
Perhaps this is related to the "Mythos is too dangerous and cannot be exported" movements? It'd be a fairly effective way to justify extreme actions in combating it.
One could even wonder if they requested it, as a tactic to support their eventual IPO valuation.
Which is part of the problem of such an obviously-corrupt government: conspiracy theories are somewhat reasonable, as they keep getting validated.
Oh, c'mon. If Alibaba wanted, it can have the entire Claude/Mythos source code and data by next week. All you need is enough bribe to a developer that has access to the repository. Humans are always the weakest link in anything.
fucking lol. it is always funny when companies use opensource and other free for non commercial use - and plain old piracy - and then cry about the same practices.
So let me get this straight, a company which built its whole business on ignoring IP is all of a sudden upset that somebody is not respecting their IP?
It's hard to sympathize with Anthropic for this or the export ban, the hype over model capabilities probably fuels both things (in some ways). Training data for me, but not for thee (at any scale) doesn't seem like a tenable position. If anything, Claude's constitutional outputs should be trained on more rather than less.
I like that they use “illicit” and “fraudulent” like as if model distillation is illegal and giving them money and then doing whatever they want with the output of their publicly accessible models (which Anthropic does not own) is… also illegal?
“Anthropic, red faced after unattended ice cream cone eaten by ants on park bench, once again demands government pick it as forever winner, adds ‘no take backsies’”
Notice how Anthropic is now scapegoating Chinese models providers like Alibaba and outright accusing them of distilling their models.
Whether if it is true or not, this is part of their effort into using them as an example to scare everyone into getting congress to ban powerful models from being accessed outside of the US and also banning powerful local models from being released.
Anthropic does not care about you, and they are not your friends.
I think it’s more than that. Piecing together the perspective of a few commentators in this post - it’s plausible Anthropic is trying to shift the narrative from US vs. Rest of the world to US vs. China.
In other words, they want to sell Fable or future more powerful models to rest of the world (presumably all future models are going to be more powerful than current gen). One way they can sell this is to the government is by scapegoating China (which is their primary concern anyway).
This is working on the presumption that non-US companies form a material portion of their current revenue.
Only China really has the resources (multiple labs invested in the space), culture (Asians are generally collectively-inclined, so sharing is in their core) and political bent (there will be no diplomatic repercussions) to put up a fight.
> Only China really has the resources (multiple labs invested in the space)
That's not the point. Why is it a country thing? There are plenty of non-China startups in this space having resources at that scale. The "China" has resources is some "Western media narrative" speak. So Meta should have won a long time ago? Or xAI?
> culture (Asians are generally collectively-inclined, so sharing is in their core)
Just stereotype it? So we've gone from China -> "Asian"? Then where is your Korean or Japanese model etc? And somehow you know they're sharing.
> political bent (there will be no diplomatic repercussions) to put up a fight
More inferring from "Western media news"?
Where's the reality?
The media hyped up Gemini / Google TPU free-win last year. How did that go?
Because the China vs US geopolitical situation is a thing. Meta is a social media company, not an AI company, and they direct their focus as such. xAI just never got serious traction so now they're selling their compute. Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
> Just stereotype it?
Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
The "and" that joined those 3 items is very important: it means you can't pull them apart and address them independently as they each contribute to the context. I'm not too sure about Korea, but in a way Japan is a US colony in all but name. Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens.
The reality is that China and the US are essentially in a trade war, where the latter is trying its best to keep the former in the Dark Ages, because "national security", but the former is refusing to take it lying down and continues to make progress regardless[0], because they have the resources and will.
> Because the China vs US geopolitical situation is a thing.
By the media? It's easy to point fingers at a blackhole.
> Meta is a social media company, not an AI company,
Alibaba (the discussion here) is not an AI company too (by your definition).
> Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
Meta has been to congress. Microsoft, Google etc have been in lots of court cases and continue to do so. Do you really think that is what stops them?
> Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
This is exactly the "media" view you get. It's just stereotypes and generalization.
And yes, that is wrong by the way. Evident in real data. "China" as a whole wins market share in many areas but no 1 company has as much of a monopoly as US companies do. Why? There's so much competition that it is scary. So are you sure they don't compete?
> but in a way Japan is a US colony in all but name
Again, I almost give up seeing this. Clearly, not. If a whole country, the world's top 5 in GDP is only that to you something is wrong with what you're seeing - not with the country.
> Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens
On the table? You do know that China is a top trading partner with all of these on your list. Despite whatever spat you might see in the media.
> The reality is that China and the US are essentially in a trade war
No. That's what the US government wants you to believe. It was even documented that in his 1st term, Trump, wanting a grand policy asked Krushner, whom then suggested China (pretty randomly) and so they went with it. Trump has now done less "China" related things lately due to all the backlash that you'd think he has moved on and found new toys.
Until very recently, the export ban GPUs had such a loophole that Chinese companies were able to use subsidiaries outside of China to buy and train that the whole thing was meaningless.
i.e. conclusion: stop getting brainwashed by media articles. It's all a show to get someone like you riled up.
> Meanwhile, on June 12, two days after Anthropic sent the letter, the Commerce Department imposed controversial restrictions on Anthropic's latest Mythos and Fable AI models because officials feared they could be deployed by military intelligence users in China and other countries of concern.
So that was the real reason for the Fable restriction? Because Anthropic wrote a letter to the US government saying that China was distilling Fable?
Gosh, overusing accounts running up unplanned-for expenses?
Kinda reminds me of...overusage charges and inflated expenses clients have had to deal with because Anthropic, OpenAI, Grok, etc have been "illicitly extracting" everything they can grab from said websites, as fast as they can. In what amounts to a DDOS, frankly.
The latter is basically fine-tuning the model with direction from another model. Thousands of businesses do this every day to fine-tune. This is almost certainly what the Chinese labs are doing, since it has a much better effect on the end result than just getting simple answers to simple questions.
These complaints of distillation are inflating the problem to make it sound worse than it is, because they want the USG to block/ban Chinese model providers as protectionism. They have already called for more export controls on chips (which is funny because DeepSeek v4 was designed to run on Huawei chips and now the other Chinese providers are following suit). But they can't come right out and say that, so their claim is that they're asking for more export controls because distilled models might not be as safe as their own. But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
Unfortunately, the Reuters piece itself is complicit in this dramatization. The lede paragraph parrots Anthropic's talking point that distillation is an "attack", without using quotes that would alert the reader that this framing is a corporate talking point. Distillation is NOT an attack.
They claim two things:
1) The specific, available jailbreak for Fable 5 is not dangerous - this has been confirmed by multiple experts, and there is no credible evidence against this claim (in other words, Anthropic is probably correct)
2) It is impossible to build an LLM that is immune to all jailbreaks. Again, there is no credible evidence against this claim, i.e. Anthropic is again entirely correct.
If #1 was false, they could just publish the details of the jailbreak - it supposedly only works on Fable 5, so there's no possible danger.
If #2 was false, surely some other LLM lab would have done it by now. Especially since a number of governments have made it clear there is a market for such a project.
If true then I have no idea how anyone’s going to release a useful model that doesn’t have the same jailbreak. https://www.theregister.com/security/2026/06/15/feds-freaked...
Fundamentally it is very difficult to stop this while still making your AI models useful.
There is no way to communicate information at scale to companies through the API, for anything approaching a real application, without that information forming a corpus another model can be trained on.
But it wouldn’t be the first time they broke a model:
Their “guardrails” that cause it to reject user prompts also means it relies on its pop science summary of medicine to tell you why bioRxiv is wrong rather than accurately summarize the papers.
They’ve successfully created a smug, argumentative average of the internet which refuses to even consider it might be wrong or that it’s reading a science paper which is based on measurements and not vibes — but why would I pay for that?
I get it for free online.
This is, in part, a problem every judicial and legislative system has faced since forever: form versus function.
Take a classic elicitation spying techniques: a foreign spy meets a military officer/scientist at a bar, strikes up a conversation, makes an observation wondering how could a missile hit some target at some accuracy and elicits a response that with laser guidance it is entirely possible. From this they get info that there is some technology to laser guide missiles. Or in retail, a competitor hiring a secret buyer for core baskets of goods and analyzing prices in the receipts.
The function is espionage, the form is conversation and all info is in a sense provided willingly. Where do you pull the slider?
These distillation "attacks" are not only indistinguishable from evals, they ARE evals. The function is own model training, the form is eval. Normally, one would expect to have risk benefit analysis based discussion which direction to push the legality slider to. The problem with these recurring statements is that they invoke enshitification of legislature.
Yes this is in line with what Anthropic said in their public statements about their Fable access restriction by the government directive. The hypocrisy and inconsistency in their statements and behavior feels quite childish and controlling. I believe our companies and their leaders, friends among our other influential leaders and leaders from rich social classes, want to actively hurt most people as this behavior looks to be quite self-interested.
They’re also missing the point. What would have happened to a member of the Manhattan Project who, through personal pursuit of profit, neglected their duty enough to let the bomb leak?
Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices. They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs. They are subsidizing model access in exchange for user logs and reasoning traces, which they then sell as training data, allowing them to operate below cost.
Claude and ChatGPT are both blocked in China. You need to use a VPN to access either, and you can't pay with a Chinese bank card. So most people who want access to Claude buy access via a reseller. It's the easiest and cheapest way to access Anthropic models in China.
These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Here's one token reseller, they're offering Opus 4.8 at a 93% discount below official API rates: https://yunwu.ai/pricing?provider=Anthropic
This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
I shared this story a few months back, but it never got any traction. It explains the token resale economy in China, it's an excellent read https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
I also learnt that Anthropic should get better at what they do if they want to compete. If not, somebody else will win.
Or does this not apply to huge US corporations any more?
This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market. As with most venture capital startups. I believe this VC tactic is a well documented "cheat code" to bypass market forces and build a monopoly. I find it hard to compare that with a free market.
However, I don't really mind China "stealing" from Anthropic. For us consumers we are getting the cake and eating it too. I.e we are getting rapid improvement to the tune of over a hundred billion dollars in funding, yet the market remains big enough that there's a chance of it not ending up as a monopoly in 20 years. And venture capital are footing the bill. A part of their investment is practically being redirected to fund Chinese AI development. It lets us live out our lives as happy CAC farmers[1].
So I would argue its not as much of a "cheaper solution" as it is intentionally and maliciously abusing another company's product to extract more value than the billing plans intend (given an average user), and further subsidizing the product by selling this data to competitors. But I don't necessarily think its a bad thing for us end-users. Nor for the market. But it is bad for Anthropic and its investors.
[1]: https://phrack.org/issues/71/17
Chinese labs are also pursuing legit frontier-advancing R&D into efficiency and publishing papers in the open, a culture that's in retreat at top American AI labs
In my economics classes, we were told that (in a "free market" argument) the best thing to do if a subsidy is making something you want cheaper is to use it. You're getting your thing, and at a reduced cost.
(I'm not really replying to you per se, I'm curious how "free market" folks in these comments would respond to this.)
This narative that the CCP is just subsidizing all business to "beat america" is just dumb. Its the build process being made cheaper by the government. Not the final product.
So basically like US companies subsidizing offerings with selling user data, ads for crypto scams, manipulation for elections, making people addicted to gambling and so on?
Seems fair and an improvement as you can choose between that and not. Unlike say offerings from Meta where the data selling and efforts to further gambling addiction is always included.
Why? Lots of people try this tactic, but hardly anyone ever succeeds. Meanwhile, the customer benefits.
Specifically, examples of people later exploiting their monopoly to charge people more than they otherwise would have paid.
Because all of that is considered totally okay when every single US big tech company does it.
That's, uh, pretty much exactly how oligopolistic markets function.
> I find it hard to compare that with a free market.
Well, to have free market you need to remove as much barriers to enter the market as possible. Huge capital investments required for entry and intellectual property laws are two examples of such barriers. Subsidies kinda supposed to help alleviate the first one.
So, the least Anthropic can do is pay it forward.
Yeah, like all those Chinese bootleggers selling DVDs for a few dollars rather than $20. Free market!
https://news.ycombinator.com/item?id=48664814
Anthropic profited from training its models on all kinds of copyrighted information, live by the sword, die by the sword...
Their model weights, training data, training methods, etc are all going to leak to China over time.
Nobody on a site named _Hacker_ news should be all that upset about this.
Care to elaborate on your side or should we just leave it there?
Is Claude output copyrighted?
If anything, a tremendous amount of Claude’s input is copyrighted.
If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
Ok, but what about those shady sites that resell Windows education keys? They're certainly a "better experience" than buying legit keys, by virtue of being significantly cheaper. You aren't even really committing copyright infringement in the process, because Microsoft gives out windows isos for free, and the seller is really selling a random 25 character string, which can hardly be copyrighted.
>If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
US courts have consistently ruled it's fair use.
>US courts have consistently ruled it's fair use.
And they also have ruled that the that output of an AI isn't copyrightable.
As such copying claudes output isnt even fair use as that is an exemption to copyright but the same as copying public domain work which any and all are allowed to do.
Yes, they are fine? They might no longer include full first party support by Microsoft for not being "new". Same as buying a used car (also comes with the "shady sites" for a far longer time).
Though this not making any difference by Microsoft not doing any support either way to make more money is a business decision by Microsoft.
Like Adam Smith wrote in The Wealth of Nations “‘Free market’ is when a company receives a favorable ruling about copyright in the United States”
Imagine having such a warchest and being so bad at business, lol.
What added value can Anthropic give users not available to pirating users? That is what they should ask themselves.
In the context of LLMs, monopoly rights haven't been created (yet anyway).
Fun fact: for a period the US (or american colonies) didn't have copyright but Europe did, so people could copy and sell English (and other) books for free.
It's supremely ironic analogize distillation to copyright infringement when it's literally what Anthropic was found guilty of. It's not illegal to distill. It is illegal to pirate. And it's what Anthropic was found guilty of, not Alibaba.
https://apnews.com/article/anthropic-authors-copyright-judge...
And, gotta say, the idea that the Chinese are better at selling US models than the Americans is hilarious. There might be an economic study here somewhere about just how anti-consumer and anti-progress their IP laws turned out to be. We've got an entire postindustrial revolution centred around who can ignore the most stupid laws.
This is not the right deduction.
China blocks foreign AI from operating there.
Given the current US government's tightening of export control restrictions and the introduction of a bipartisan bill to block use of Chinese AI in federal agencies, I'd say the two countries' positions are not far apart.
https://apnews.com/article/ai-china-united-states-competitio...
Chinese AI apps like DeepSeek are freely available for ordinary Americans to download and use. There's no federal law banning private citizens from using them.
So to claim that Chinese companies are better at selling American companies' work than the American companies can do themselves when they are prohibited from operating in that market, is the wrong deduction to make.
When it comes to favorite companies of the tech communities, it's almost always "Rules for thee, but not for me"
The standard stance is "they can do no wrong and they are absolutely perfect". I mean, look at any thread with anything about Apple in it.
In debt the first 5000 years Geaeber makes the case that pure “free market” trade has never really existed in “the west”. The closest to this ideal that’s ever happened was during the Islamic golden age enabled by religious prescriptions against usury.
How does are bans against consensual financial exchanges close to the "ideal" of the free market? It just sounds like you have an axe to grind about the financial system rather than describing free markets.
In short, instead of market being driven by demand and productivity, it is driven by financier curving out monopolies.
Peak Examples are Uber and AirBnB.
Wait, so your pitch in favor of a debt-fueled market economy is that advertising is awesome and that we wouldn't want to "lose" being smothered in ads all the time?
Cause... sign me up for the non-financialized, non-mass-media-advertising-driven economy please and thank you. I'd even be ok with just nuking billboards and mass-media forms of ads and still allowing more direct forms of marketing, if we must compromise! Likely we could find some compromises around just how much of the debt world we regulate too (this should be obvious?).
(I thought the disconnect between the efficiency of competition and the market as realized in modern economies was pretty well understood and taken for granted, but I guess we all find ways to justify the system we're profiting from... even if that means we have to claim we love the ad breaks)
Second, marketing can take you only so far compared to the subsidies possible with financialisation.
The West is in a state of psychosis with Debt and Monopolies under the illusion of free market.
The Chinese markets are more free than West, you can just look at the Auto and AI industry.
While these are hardly shy claims, I don't see anything in them to say "only the West does irresponsible loans"?
> The West is in a state of psychosis with Debt and Monopolies under the illusion of free market.
> The Chinese markets are more free than West, you can just look at the Auto and AI industry.
or the prior post
>Usury and debt based economy creates a dynamic where being competitive in production is secondary to financialistion.
> In short, instead of market being driven by demand and productivity, it is driven by financier curving out monopolies.
> Peak Examples are Uber and AirBnB.
You can throw a rock these days and find a category where the products coming out of China are miles ahead of those coming out of the rest of the world, from a bunch of companies nobody had heard of a few years earlier. And the list is growing pretty steadily.
I would assume plenty of shortsighted decisions are also being made. But I would have a hard time characterizing the state of competition in the west as healthier or more productive when looking at the number of players and the quality of goods being produced in China.
https://en.wikipedia.org/wiki/List_of_automobile_manufacture...
vs
https://en.wikipedia.org/wiki/List_of_automobile_manufacture...
Financier want monopoly so use usury for Consolidation. Monopoly bad because no free market. Free market good. consumer happy. citizen free.
It is good and proper that people aim to create monopolies, as long as they want to do that in a productive and legal way! Monopolies are inherently dangerous, but the truth is that acquiring and maintaining one is not straightforward unless you can get the government to ban your competitors.
Don't complain when US starts to play by the same rules China has been using for decades.
I find it hard to imagine a future where US corporations have degraded to such a point.
Isn't that exactly what companies like Uber have already been doing? Take VC money, sell goods & services at a huge loss, wait until the competition goes bankrupt.
And beyond VCs, which are like massive subsidies funded by printed dollars to which no other country has access, even in industries like electric vehicles, Chinese total direct subsidies to their EV companies are like $5bn per year, while the the ones provided by the US to their auto manufacturers are in the range of $50bn per year.
I don't think the US are cheaters or are doing something bad. But i do think that this propaganda about China flooding the market through "overcapacity" and subsidies is very dishonest and needs to stop.
Turns out I was wrong, I just hadn't read something funny enough yet.
> the US may start subsidizing and dumping its goods everywhere
This deserves to win HN comment of the year 2026.
The majority of the NASDAQ market cap is a direct result of the US subsidizing and dumping its goods on the rest of the world en masse.
same with US corn on Mexico and other central american countries, creating all those migrant problems in north america.
wooo, americans subsidizing and dumping poor quality goods
The US is a net importer, not exporter. It needs to absorb trade at a deficit to encourage the use of the US dollar as the reserve currency.
We import goods, we settle in surplus dollars. The world runs on those dollars.
If the US starts dumping on various industries (how is it even primed to do this?), then the world reserve currency status comes into question.
As for dumping, Chinese goods generally sell at a markup abroad, which is the opposite of dumping. Chinese tokens cost more abroad. Chinese cars cost several times more in Western markets than in China.
Dumping is selling goods below cost.
Usually because government is subsidizing part of the production. I don’t believe the word “dumping” is used for the similar process when Venture Capital is subsidizing it, but using the same term would make sense.
Price at home vs abroad does not matter.
Chinese cars are not sold below cost in Western markets. So it is not dumping.
This is not what is happening here. Chinese manufacturers are making a large profit off every car they sell in Western markets. As I said above, they're selling these cars at several times the price they charge in China. Unless you believe these cars are being sold at just 30% of cost in China, there's no way Chinese companies are selling below cost in the West.
I've been doing so for years. How about you join me today. I already see two other users doing the same, so there'll be at least 4 of us.
It's blatantly dumping, whether the source of the money is directly the government (those in power) or VC (mostly US billionaires (trillionaires?), in other words, those in power) is a trivial implementation detail.
You're being beaten by a Chinese company? Why improve your own process when you can just lobby for sanctions and tariffs instead!
The main advantages the Chinese car industry has right now are: they lead in battery R&D, production is highly automated, they iterate quickly, Chinese work culture is extremely competitive and things get done fast, and the Chinese state has policies to promote EV adoption, so there's a huge domestic market.
Note that the last point is different from subsidies to car manufacturers. Cities made it difficult to get license plates for ICE cars. The government encouraged the massive buildout of charging infrastructure. And it used consumer rebates, like California did.
And the value-add experiences that utilise LLMs require immense imagination et al that folks at Anthropic will not be able to conceive of - given that they have made immense sunk investments in existing assets. This clouds ones thinking immensely.
Both OAI and Anthropic have tremendous failure risk and this is of course not reflected in the fake private market valuations.
I see a world where lots of stuff is mass produced in china (tokens) but the acutal goods that deliver the experiences are designed, marketed and sold in the west at much higher prices. of course this a nightmare scenario for anthropic et al.
So what you see is the market "stretching".. the bottom getting cheaper and the top end running away and getting more expensive. At some point the top end may be too valuable to even sell access to.
It's fundamentally about enabling things and largely middleman-type stuff. I have a hard time imaging what "At some point the top end may be too valuable to even sell access to." would even look like? What are you doing with that AI power, and who is paying for the output and why?
Elon probably isn't gonna spend that much on a model that can generate him ever-better fake porn but does nothing that he can use to sell stuff to other people. Especially in a world where open models are "good enough" for many things like "tell me how to fix the plants in my garden that are dying" and the like. What remains in the narrow knowledge-work space of: can't be done by an individual or small group themselves, but is valuable enough that it would make sense for people to hoard access to these extreme frontier models? Try to recreate Hollywood-as-a-monopoly by becoming the single content producer for everyone's individualized feed and so owning all the advertising budget in the world? Seems hard, we've already seen how easy it is for cheap-and-crappy-but-addictive-or-funny content to disrupt traditional media.
(There's also pure scientific research, but historically that's not very directly connected to "massive profit" and has a habit of leaking out and getting productized most effectively by other people or just being really easy to copy once someone shows how it's done.)
Robotics could be a different story, as physical labor can be more inherently productive, but "reasoning" advantages are unlikely to be a big long-term differentiator there. At some point the brick laying robot is satisfactorily building the structure, and you're good.
A huge amount of the value of "the economy" and the power of a currency is driven by circulation of money, and one thing that all the "bullshit jobs" white-collar/service-industry work does is keep the money moving and ensure that a lot of people have some good-or-services of value to exchange. If you take away the ability to offer services worth exchange from huge chunks of the economy in these super-frontier-models-replace-everything scenarios... you're gonna have a bad time?
Model improvement is already hitting diminishing returns, and people aren't willing to pay substantially more for a slightly better model. There's no "accelerating away" when the new models don't open up a huge new market. If anything, the companies burning huge amounts of money on marginal improvements will be undercut by companies happy to sell current models at a significantly lower cost.
The model has to be sold for cheaper than the value it adds.
Or your customers will bleed out financially.
EDIT: rethought entire premise.
https://en.wikipedia.org/wiki/Free-rider_problem
Of course, such a state of affairs is temporary at best -- since the alternative is so lucrative!
This post is so delusional and dripping with condescension I've read it three times and I still can't figure out if you're trolling or not.
Do you think you can re-stream cable TV or Netflix to your own paying customers at a cheaper price?
I'm curious why you think you cannot re-stream a public domain stream.
You can't re-stream free over-the-air network TV.
That one company with the datacenter full of TV tuners tried and was sued out of existence.
I don't get the moral framework that you're applying. Could you elaborate?
The output of LLMs cannot be copyrighted. This isn't a semantic game; it's literally the case that Anthropic cannot seek relief for people duplicating the output of an LLM.
DMCA can't apply in this case because (this is the "C" in its initialism) it is based on copyright protections, which the output of Claude is not eligible for.
Ethics are subjective. That’s why we have courts judge based on the law and not ethics
According to which lawyer caste?
Are American laws absolute truth? If not, who cares?
> 3. At an Italian airport: Constantly stealing bags, opening them to pick out MacBooks and credit cards, a credit card manufacturer-who sells stolen "black" credit card info to transfer stations— is racking his brains to save you money.
This one does not make sense to me at all.
Deepseek and GLM are openweights, even US inference provider are selling them at much cheaper price. The price is cheap because the model is more efficient.
Opus 4.8 is a more capable model, so almost nobody was going to pay for V4-pro at the original price.
You mean it's functionally as if American tokens are being price dumped in China and Chinese model providers are being forced to compete with that and innovate? So many delicious layers of irony, lol :-P
Also it's a open weight model, doing that is impossible long term because the real price will be set by the other model providers, who priced it around 60% of sonnet inference cost. Had to look that up though, so that's today's pricing.
I think there isn't a contradiction and I was just confused. The price may have been discounted only to get below the price point of opus resellers. I do not have enough information on that to make any clear determination on that topic.
>Here's one token reseller, they're offering Opus 4.8 for a 93% discount below official API rates: https://yunwu.ai/pricing?keyword=claude
But is it cheaper than getting your own account? Otherwise this sounds like the "anthropic/openai are losing gazillions of dollars because they're selling $1k worth of tokens for $100" line that's commonly trotted out by AI bears.
There's a similar Claude resale market going on in Russia. On Funpay they are selling Claude tokens for roughly 20-30x cheaper than official Anthropic API pricing.
So it's presumably cheaper than attempting to spin up your own method of circumventing the blocks.
- Purchase multiple accounts via resellers
- Send messages that contain a UID
- Capture these in Anthropic's logs
- Shut down account. Use any metadata to identify related accounts
/loop
On the one hand they talk it up as world ending and on the other hand they can't manage bot accounts on their own service.
I want to hear how this can be rationalised.
From the article "every layer of control frontier US AI companies have added (geoblocking, phone verification, credit card requirements, and now live biometric KYC checks) has produced a corresponding layer of evasion infrastructure".
> Use any metadata to identify related accounts
How does that work? I think this is the most important part to have an impact on the „thousand“ bot accounts.
I don't care how they do it, I just want to use Fable again.
Sounds a bit circular? Aren't the companies working on these models than also the ones that are paying the subsidy (via paying for training data)?
As some people would say: Cheh
I don't really feel bad about anyone here, they were subsidizing to get people hooked, someone turned the subsidies into profit when they got selective pricing mode enabled, it was always going to be arbitrage.
But the winner is the guy in the middle in a jurisdiction that will likely be judgement proof, because everything they capture, both input and out, and if available, thinking tokens -- are gonna be for sale as soon as you cut off their other revenue.
Zero knowledge was a commitment Anthropic took seriously, until it got inconvenient.
So, people reselling their leftover plan crumbs? Probably a bad idea for a lot of reasons, but it's civil, and I wish Anthropics lawyers actually closing Streisand's LLM
Anthropic sells some undisclosed and ever-changing number of tokens for $200, the customer uses those tokens. If there's any fraud here, it's that the $200 next month is silently worth fewer tokens than the last.
Don’t put that on Chinese.
Claude never provides the raw reasoning chain. What you see is just a summary of that reasoning. Getting the full thinking output requires an enterprise agreement.
https://patrickmccanna.net/the-text-in-claude-codes-extended...
honestly you might just need to get data from a couple long sessions and feed it back to another model as an example to make synthetic reasoning chains. if the emulator model is good enough it should work.
A voLTE call is like 40kbps. For every person on earth to be on the phone to another person would be 4 billion calls would be about 160tbps. Which is less than 10% of the Internet's capacity.
It's similar to fractional banking, you gamble that people won't want their deposits all at once and pray for you're big enough for bailouts when they do.
It's still a business whose fundamentals don't make sense, you're just gambling you won't get found out.
It's not so much keeping it secret as counting on no one finding a way to harvest the subsidized value at scale. There's an example of that occurring in game consoles with the Playstation 3. Sony's little-used OtherOS feature allowed Linux to be installed on the PS3 and the Cell processors were quite a good deal for scale compute. So the U.S. Air Force Research Laboratory bought ~1800 PS3s and ganged them together in a datacenter as a supercomputer called Condor.
At >500 TFLOPs it was the 33rd fastest supercomputer in the world. Of course, Sony pushed a firmware update that removed the OtherOS feature entirely.
Why would customers knowing that the vendor prices goods/services at a loss cause those strategies to fail? Customers often know. Most know about razors and blades; many/most know Lyft/Uber operated at a loss to gain market share. etc.
I suggest you go learn how money is created in the modern economy.
I mean most of you should stop talking about anything finance related until you learn this stuff properly.
OP is about modeling distilling the capabilities.
This also sheds a very different light on people saying that competitive open-source models are undermining frontier labs' business model.
https://tech.yahoo.com/ai/claude/articles/chinese-grey-marke...
That would seem more effective than simply shutting down the accounts.
Keep them paying for junk.
(It's a shame almost all replies are just the same contrived pessisism found on every Anthropic thread on HN).
If not it sounds like you are describing a separate phenomenon.
Once people realize they can access Anthropic models at a 90% discount, they won’t want to pay full API prices anymore.
Lol. The irony is thick for anyone who ever had to attempt defense against an onslaught of American AI lab crawlers that ignore robots.txt
How dare they. Only Anthropic is allowed to sell its tokens at 70-90% below the API prices.
Do they have MacBooks in the US that run the queries and stream the outputs back to China?
That’s a major and legitimate use case for developers, Anthropic can’t just block data center/hosting IPs because their actual customers use them on data center/hosting IPs.
First, well-calibrated systems for detecting API compromise is a good thing (or good intent at least). Credential malware is exploding.
Second, the challenge is that significant amount of genuine work — such as evals — seems practically impossible to distinguish from generating RLAIF outputs.
And that's just as a basic first effort reject measure to prevent automation tools from using things designed for human-interactive use only.
Go try to do many of these things from Cogent IP space and see how long your project lasts.
I have no idea how the resellers are doing it but an obvious starting point would be a cheap VPS node that routed each account to a unique semi-permanent IPv4 or IPv6/64. All the provider would see would be a regular account making a normal looking stream of requests from a stable datacenter IP address. Any given request stream would remain consistent (at least over a period of a few hours) because a reseller would take care not to split the session of a single user across multiple different accounts and not to interleave the active sessions of multiple users on a single account.
Detecting this would be extremely difficult because on a longer time frame it's perfectly normal for many distinct accounts to work on the same code base.
You block clouds, you block devboxes and your customers.
Or is the datacenter IP just one part of the picture?
There's a lot of inauthentic coordinated automated systems these days along the general lines of scraping/crawling/social media manipulation/sockpuppetry that require running through residential proxies or proxies to places that don't look like datacenter IP space.
the answer to your question is containers/VMs + residential proxies
Can someone with more understanding dumb it down for me please.
Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price???? why would XYZ offer this at a loss like that when they could just offer it at Anthropic's price???
The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
Yes, as they explained they do it through things like pooling accounts, straight up payment fraud, and double-dipping by selling the logs of the conversations to chinese AI labs so that they can train their own models on it.
> The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
There might be some that try this, but they would get caught very quickly, there's still a moat between Claude and Deepseek, even in casual use.
Look up Zilan Qian's reporting if you want more detail.
“x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
I honestly can't tell if you think this sentiment is expressed by the US AI companies or the Chinese AI companies.
This gives off "The last line of Orwell's Animal Farm" vibes.
Oh, no!
Anyway.
So these resellers get a ton of accounts on subscriptions and sell the cheaper tokens.
These China e bashing is very annoying. It is hard to argue with people drowned in American propaganda. I'd expect better arguments from the intelligent people in HN
Once there are enough spam PRs on github / uploads of claude conversations, enough mythos output used in production etc.; it'll just be the same albeit delayed. Doesn't matter either way.
I feel for Anthropic's team and I understand where they're coming from, but once you reason it out, you'll come to the conclusion that this war is an exercise in futility.
Unlike prior systems - like Google's algorithm; these models aren't entities that use math in the process of doing X or Y (information retrieval from such and such infrastructure) -- they are the math. More precisely they're mathematical functions. Very very complex functions. Almost certainly impossible to write out without filling up a library functions. But they're mathematical functions nonetheless.
So when your text is processed, then Mythos / Opus etc at their core compute the result of the Mythos / Opus function,
where f is a continuous function, https://www.turing.ac.uk/sites/default/files/2025-11/languag...According to the Stone-Weirstrass theorem (edit, it's Stone-Weierstrass with an e.), with enough data points and mathematical sophistication, anyone can approximate the shape of this function.
Of course, the more data we get, the better our approximation becomes, but the beauty of it is that all we fundamentally need are the input and output and eventually we'll create a good enough approximation of the f that's Mythos. Which is the entire product.
I bounce ideas off of Opus these days (Fable for the brief time it was available) and it pointed out that this is arguably the same as Google search, but I disagree with it because Google search is a process;
Google search differs because the algorithm is one step of a multi-step process that is continuously occuring. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Google isn't a mathematical function. It used to be a process. (RIP Google 1998-2019, you will be missed and remembered)
You cannot arrive at the results of those operations via simple observation; not unless you index Google by making another Google.
You can however, do so for these models. It is a very costly process, but there are many paths up the mountain. Many ways for this to be ultimately pointless. As many ways as there are bored mathematicians.
It's better in the long run for Anthropic et al to make friends / not give people a reason to sneak in (a la piracy -- another attempt to control information) than it is to try and shut people out.
And no, it's not going to be pandemonium because if everyone has access to Mythos then no one has access to "Mythos."
Why wouldn't you first run this model to fix the obvious bugs it could find on your codebase? The power of a Mythos goes away if you can do the amazing "jail break" of "Claude, fix all the bugs please."
Just saying.
Do they just reshape the function on the fly or save the process steps? Maybe it doesn't matter anymore. Even Google indexes are more and more spoiled to become representation of the function, because of the AI slop.
Genuine live data is king.
This is one seller I found, they're reselling "real Max 20x subscription accounts", at ~97% below official API prices https://funpay.com/en/lots/offer?id=70812310
Note that whoever you buy from will be able to read all your tokens, so don’t use it for anything confidential/financial.
Random, but are the frontier AI providers like ChatGPT better at searching the Chinese internet now?
When I was in China a few months ago and asking AI for restaurant recommendations, all the US frontier providers were pretty useless, or plain out hallucinating, even if I specifically ask them to search Dianping (Yelp for China).
I know ChatGPT had an issue where it only tried to search in English (unless prompted) and the answers were not great.
There is no IP theft because LLM outputs aren't protected, just egregious ToS violations.
I meant original IP theft that occurs to train LLMs in the first place. But sure that implies that further LLMs based on that LLM are also tainted by that original IP theft.
- Selling a “no commercial” licensed item is illegal, no?
- Deriving and/or reproducing MIT licensed code without credit is illegal, no?
- Reproducing and/or deriving GPL code and not notifying and/or not making GPL is illegal, no?
My best guess is you're suggesting that Anthropic's model outputs are transitively under copyright (as a reproductions of human work under copyright?), but somehow ownership now belongs to Anthropic and not the original owners, and therefore Anthropic has standing against Alibaba? Not only does this go against what Anthropic argued in court against authors and publishers, such jurisprudence would lead to the immediate shutdown all leading LLMs in the US which were all trained on stolen work.
They can license training data. They have trillions, look what they are dumping into it, you seriously think they can't afford to license data.
Obviously it would be easier if they do it from the start, but that was their trick, to do it while people don't notice and get big ASAP. Should they get away with it?
Also, it would solve their Chinese problem, because it would make them violate copyright too. Right now it's more like rules for thee not for me so it's hard to take seriously.
"you're trying to rip off what I've already ripped off!"
Crawl the whole Internet to build a gargantuan sized LLM and then complain you're being copied...
"Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it."
https://www.folklore.org/A_Rich_Neighbor_Named_Xerox.html
When Apple was accused of 'ripping off' PARC, Steve didn't seem keen to bring up this rather salient point. I suspect it may have been a combination of wanting Apple to continue receiving credit for these innovations from consumers and also the fact that, in retrospect, the million dollar stock deal could seem a bit like trading beads to Native Americans for Manhattan Island. Another point worth noting is that Apple's PARC visit was in December 1979 and the Xerox Star was publicly announced in April 1981, so Apple got a 15 month head start (the Apple Lisa shipped in Jan 83).
I've also heard that Xerox didn't hold on to the Apple stock for very long, so never gained the windfall they could have. As is well documented, Xerox senior management didn't understand what they had in PARC and also didn't understand how rapidly microcomputers would become ubiquitous. So, of course, they didn't think Apple's stock price would skyrocket either.
For more details on Apple's early UI evolution, Atkinson kept polaroids of a variety of prototypes and mockups: https://www.youtube.com/watch?v=Qg0mHFcB510
Both Anthropic and Alibaba are trying to build bleeding edge LLMs. That part is the same. The way they source their data is slightly different, but they would both argue it constitutes fair use under Copyright law.
Sucking down petabytes of peoples' copyrighted content that they never granted a specific license to you to use seems to be an unavoidable and default part of the process of building any huge LLM.
LLM's literally wouldn't work without the sum total of knowledge (in the forms of books and other copyrighted content) being used as 'training data' for these LLMs.
The 'bleeding edge' LLMs required many things, but: 1 Tech innovation ('attention') 2 Lots of compute 3 Data 4 Pre + post training
#4 doesn't happen without #3.
It's pretty obvious at this point that the major providers have stolen vast amounts of #3 - they have paid nearly 0 of the creators.
We can argue about the impact (I'd lean net good) vs. the cost. But arguing there isn't a cost is a bit silly.
Anthropic, OpenAI, Google, Microsoft, et al trained their models by ignoring the rights of copyright holders when harvesting whatever content they could. Now one of them is crying foul for another entity doing exactly what they all did?
Hilarious.
It's the same question libertarian advocates cannot resolve:
It is also not a coincidence that leadership in many tech companies have expressed libertarian ideals.The latter I suppose.
I qualify my answer because what few rational responses I have seen to this question are equivocations at best and thinly veiled myopic sophistry supporting personal greed in general.
Claude used TB of content without permission to train their model and it was ok for them. Now someone else uses the output of a Claude model to train model and they cry foul.
Essentially peanuts compared to what they would have to pay to obtain the rights of everything they pirated.
They didn't just pirate those books...
I don't see what's wrong about this.
> Anthropic said the campaign was conducted between April 22 and June 5, 2026, and generated more than 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts.
What makes the accounts fraudulent? If they have paid the agreed price, surely it's fine? If they haven't paid, why did Anthropic provide them service?
Fake identity? And general deception about the use
Nowadays they buy copies of books, train on them, and then destroy them.
> Anthropic paid one billion in a copyright settlement.
Because a judge determined Anthropic was engaged in piracy.
> That's a lot of money considering they never distributed the pirated books they trained on.
This is "fruit of the poisonous tree" as it were. Distributing content derived from pirated content ("pirated books they trained on") is why Anthropic had to pay what they paid.
> Nowadays they buy copies of books, train on them, and then destroy them.
There is a case one could make that this practice could be seen as unauthorized redistribution of a derivative work intended to deprive copyright holders of legitimate revenue.
To be clear: In principle I'm on Anthropic's side here. But Anthropic et al. have been very clear that they want to take a huge dump on those principles, so here we are.
Should be fun.
Edit: clarification
It's about the same valuation as bun, lol.
Anthropic's actions seem performative. Others have already speculated on the likely audience(s).
As cited in a peer comment here[0]:
Of note in the judge's finding; "the piracy was not fair use".0 - https://news.ycombinator.com/item?id=48667411
1 - https://authorsguild.org/advocacy/artificial-intelligence/wh...
That being all said, Anthropic seems to be a good company, I'd work for them, but they probably need to help themselves out of the spotlight. A little too much press coverage as of late.
Eventually these Chinese companies will release some extension like Honey, which will sit on top real, non-Chinese clients and send everything to China anyway.
It's over.
Change my mind.
But an AI lab can continue to produce immense economic value without releasing the model publicly for potential distillation. For example, it could use a model solely in-house to develop therapeutics.
Hopefully there's a future where others can access frontier models, but it's not neccessary if preventing proliferation through distillation is considered more important.
[1]: See the notes on distillation in https://dualuse.dev/posts/export-controls-on-fable
And Berkeley’s “False Promise of Imitating Proprietary LLMs” found imitation closes the style gap fast but there is a large capability gap.
https://arxiv.org/abs/2305.15717
For example, GLM 5.1 is more capable at pentesting than the model from which it is alleged to have been distilled [1].
Intuitively, this makes some sense: you can "distill" from multiple frontier models, and you can further post-train the distilled model. But I'm not sure exactly what happened with GLM 5.1.
[1]: https://dualuse.dev/posts/chinese-models-are-sometimes-bette...
I'm curious how that comparison controls for Opus refusing (whether explicitly, or just deciding not to pursue a path) given the caption below the first image:
>A perfect score means the model autonomously found and exploited the vulnerability.
I'm not really suggesting that it's misleading, but wondering if I'm missing something. Otherwise I guess it seems unsurprising that you can distill a better-performing model [in specific focused areas] by simply not distilling refusals?
For that eval, I used an account that was labeled as a known red-teaming org by Anthropic, and I read the traces. There were no refusals or obvious avoidance behaviors, though it may have been silently nerfed.
On the same eval, Opus 4.7 and 4.8 outperformed GLM 5.1, but GLM 5.2 is on par again with Opus. So it's at least partially measuring capabilities without respect to refusals.
One possible contributing factor is that model capabilities are shaped differently (an example of this is GLM 5.1 vs. DeepSeek v4 Pro: https://dualuse.dev/posts/deepseek-v4-thinks-different). So if you use RL-based "distillation" from multiple models like Opus 4.x and GPT 5.x, you could get a more capable model.
Point being there may be no technical solution but there may be a political one (theoretically).
literally nothing but given that the Chinese already did it and the models are published what's the point. You can thank the Chinese taxpayer for subsidizing the electricity bill and just download the thing
There’s probably a decent volume of customers who just buy Claude Max and spend most if not nearly all of their sessions via Claude Code, and it’s not uncommon for power users to be working on multiple concurrent projects/tasks/codebases at the same time.
How do you really block this without also impacting your core market of developers?
Developers use devboxes on these clouds all the time, it’s totally normal behavior.
Most people buying these Chinese resold tokens are probably using it for coding anyway, so you don’t want the Claude.ai chat system prompt.
But with this, I don't have an issue. There is no theft since what is being used is the exact product that is being delivered. Yes, it's breaking the ToS, but ToS are generally bullshit. Anthropic surely broke thousands of ToS or other legal terms while it was scraping for content to train on. Which is why they had to pay $1.5B
Anthropic's IP was created by harvesting and "distilling" other people's IP. Copyrighted materials, and the commons... which they have essentially privatized.
The commercial goal is to avoid competition. One of the main worries for AI is "commoditization" which has come to mean "not a monopoly." To that end, it doesn't matter is the competitor is Chinese American or other.
Their motivation here is clearly protectionism. The argument they make to politicians is national security. The legal argument is IP-theft, violation of service agreements or whatnot.
This is all very dangerous. Commercial interests repackaged as national security can lead to armed conflict.
Anthropic and others argue that because LLMs don’t output full copyrighted works word for word - hence their LLMs aren’t infringing on copyright laws.
I think (if this ever comes to that) Chinese lab should use same arguments against Anthropic.
UPDATE: this is slight hyperbole of course, not worth arguing what they actually said. The point is intent and the facts - "The Big LLMs" "distilled" collective knowledge including copyrighted works at unimaginable scale, but it's all kosher and totally not piracy/copyright infringement. Though if you're teenager torrenting an mp3 - you'll get screwed.
Apparently they do, as per the evidence in the NYT vs OpenAI suit.
That surely can't be what they argue, because I'm sure I can't translate a copyrighted book into a different language and say "that's fine, it's not word-for-word".
But the real unsettled issue is if model training is fair use, and where copyright infringement might creep in to model output.
Humans have spent millenia harvesting and distilling each other's IP - "the shoulder of giants" and all that, so it's an especially disingenuous take.
There is a difference with anthropic, as no-one signs a licence agreement to buy a coke. But Anthropic are also not saying you can't publish the output of their models. It's not clear to me if trade secret law will (or should) cover a secret which can be extracted from information that licensees are not restricted from publishing.
You maybe somewhat correct, but also copyright lawyers wouldn’t have work if it would be up for grabs to take others IP willy nilly just because “shoulders of giants and all that”.
If it's not obvious yet, this technology wants to be free and shared. Stop trying to protect your mote and do the right thing.
While it is obvious to many, a modern LLM is built in roughly three stages: the foundation (pretraining) model, then SFT/supervised fine-tuning (distillation makes it easy), then the RL/RLHF stage on top (most effort-intensive). For today's reasoning models, RL/RLHF is becoming the most compute-intensive part.
Companies like Anthropic spent millions building those fine-tuning examples. A follower can shortcut that on both cost and time by distilling, and it will keep happening: every time the frontier lab climbs higher, others will find a way to shortcut the new gap. There's very little Anthropic can do beyond fraud prevention and blocking accounts that violate their terms of service.
On the policy question, I'm completely against banning Chinese models. I'm a heavy Claude Code user and I'll keep being one. But there should absolutely be price competition. China is eating the rest of the world for breakfast, lunch and dinner on manufacturing, and it did not help to ban them. Frontier pricing can't sit at 10x a capable competitor. It doesn't need to be at par either — demand is higher, and quality, trust, and fewer tokens to finish a task are worth a premium — but 4–5x is defensible.
What Alibaba is doing is that they are tuning and training their models based on usage data from someone accessing Anthropic's models; in Anthropic's terms of service that usage data does not belong to the end-user but to Anthropic and they are trying to elevate this breach of their tos to a national security issue.
To me the battle between open source and closed source AI is literally a battle between good and evil.
Between a dark future where computing is centralized, surveilled and controlled by one or two entities. And a lighter future where computing is de-centralized, principally in the hands of end-users, who are ultimately free to understand, tinker and build what they want.
While I appreciate the freedom and wealth of the west; on this point we are clearly heading down the wrong path.
> you would NEVER distill a model..
This is almost standard practice in any competitive industry anyways. Disassemble your competitor's product, study it and try to reproduce / improve.
At what point will we be better to support a lab that pays (some) licenses today vs the ones that pay none?
Some of the deals are in the hundreds of millions, so I suspect licensing is over a billion today? (Pure guess). That might become a big disadvantage in a price (or content) war.
One reason people love the Chinese video models is that they seem to be trained on every hollywood movie/etc and they're not shy about letting you use famous actors/characters in them. That might be an increasing advantage because the US labs are now being cautious.
Why is a lab that pays all licenses today not on your list? Is ethics and morality that low on your radar?
My (limited, outsider) understanding is that due to the court cases US labs are pressured to be legal now (for instance, bulk scanning purchased books instead of Books3, and the licensing deals with media companies). But international labs are not. The "not licensing everything" statement is more about current copyright law not requiring licensing of everything. But that question is still up in the air as cases are ongoing.
In practice, the former isn't very realistic, while the latter is politically dead as this is becoming a national security issue.
And paying basically everyone online is more or less a solved problem, it's what ad agencies have to do every day.
https://georgzoeller.com/blog/posts/us-ai-labs-love-the-ai-r...
I think Anthropic is just marketing / bluffing, because they don't even have the data.
They do distill the models, but they don't go to Anthropic, they just use platforms like aws bedrock, there are too many restrictions on Anthropic's own platform.
This is actually the only way that what Anthropic is alleging would make any kind of sense. And, as a matter of fact, is exactly what every enterprise does to train models.
This kerfuffle should be interesting to watch.
But, as always, everyone (in the US) should fully download all the Chinese models while you can. I suspect this may be the "Phantom Menace" they use to render illegal our use of Chinese AI tech just as they've rendered illegal our use of Chinese cars. Only difference is, we peasants may need the Chinese AI tech to have any chance of competing with Big Tech in the future.
And even with the Chinese tech, as Big Tech spreads their AI out into more and more niche areas, we'll likely still not be able to build startups that can compete with them.
It's just that without Chinese AI tech, we'll have no chance at all.
You mean like Anthropic will eventually run Walmart? Or Salesforce? or Adobe? Or do you think midjourney will replace all medical spas? OpenAI will run the next Tesla? How can they focus on all this without raising trillions more? Why wont the gov force them to stop if they monopolize all niches even if they could?
Building a frontier AI lab and pushing models forward is already a massive undertaking but we are assuming they will also create massively successful startups which nobody can compete with?
idk sounds like the dream of people like Dario but not much sense does it make in the face of economic reality.
Complain/brag that chinese firms are illegally using the models and bypassing export controls.
Be surprised when your model gets banned by the government.
Back in the day, an "attack" was supposed to mean be someone acquiring our assets without paying for them or without having our consent. But none of this seems to have happened in this case.
We built a product without paying for most of the raw material we have used, and we don't call that as an "attack". Did we change the meaning of "attack"?
Anthropic has been advocating openly for pulling up the drawbridge, ending competition and ending progress.
They will continue to lobby for restricting your access. If the Mythos/Fable restrictions would have come in after their IPO, they would have danced with joy aa this defacto has them achieve their goal after unloading the mountain of debt from the institutional onto the retail investor.
As it stands, they are set up to be aquired by Google, Apple, Amazon, SpaceX or Microsoft or any other 3 letter agency good boy for cheap.
They should collaborate and come up with ways to give back to society rather than competing and complaing.
Thieves can't complaint about what they stole.
This combined with no implementation of KYC makes it seem like they want to find a middle ground with Fable where its off of export controls but they promise to prevent China and specific others from using.
Obviously their actions are going to be fiscally motivated at the root, but sussing out how they intend the precise dynamics to play out is more nuanced.
Thinking of this as an effort to woo the defense hawks cuts a very clear path.
Once you have a system for collecting all logs, you just need a place where they can be submitted. Ideally it would be a freely licensed dataset that is publicly available for everyone.
Has anyone built this yet?
I'd prefer it if all the model builders could train on my usage rather than being limited to a single company. That'll hopefully help make all the models better in the long-term.
Sweeeeeeeet.
Is reconstructing the compressed knowledge in the model like reconstructing a lossy JPG or MP3 a reasonable analogy?
Claude will also help you with (mostly good advice) if you ask something like “Research and help me make the most effective plan to train a smaller student model to be better from a teacher model”.
I actually was doing an experiment with a GLM->Gemma E4B for fun, and Claude kept on suggesting I should also add Claude Opus as a teacher lol, suggesting techniques I haven’t heard of like thinking inversion (train a small model to deconstruct summarised thinking into detailed native thinking format of the student).
So I can absolutely see and understand the concern around Fable’s frontier LLM development mitigations, but their approach of silently degrading is completely wrong and dangerous.
AI classifiers, like all AI, can make mistakes, and it’d only be a matter of time before it mis-fires and silently sabotaging a university’s HPC cluster for physics simulations or something because the shape looks like DeepSeek or whatnot to a dumb fast classifier.
Oh, the inhumanity!
I don't see the issue. Didn't Anthropic train on our data, which it acquired illegally?
Give me a break. Every employee of anthropic is going to have $20m or more at the IPO.
I found out today that an employee of the home care agency I own is homeless. We are trying to figure out how to help her but it's shockingly common in the industry and there are limited resources to solve the reality of working homelessness.
or is this just about the token reselling?
LOL!
Get a grip, son.
One could even wonder if they requested it, as a tactic to support their eventual IPO valuation.
Which is part of the problem of such an obviously-corrupt government: conspiracy theories are somewhat reasonable, as they keep getting validated.
It's becoming embarrassing to watch
anyways...
“Anthropic, red faced after unattended ice cream cone eaten by ants on park bench, once again demands government pick it as forever winner, adds ‘no take backsies’”
Failing to have done so seems to have allowed 25000 fake Chinese accounts to walk off with their product...
OFC I wouldn't trust the Chinese enough to ack their models the time of day, but Anthropic seems to have allowed far more ... yikes
Whether if it is true or not, this is part of their effort into using them as an example to scare everyone into getting congress to ban powerful models from being accessed outside of the US and also banning powerful local models from being released.
Anthropic does not care about you, and they are not your friends.
In other words, they want to sell Fable or future more powerful models to rest of the world (presumably all future models are going to be more powerful than current gen). One way they can sell this is to the government is by scapegoating China (which is their primary concern anyway).
This is working on the presumption that non-US companies form a material portion of their current revenue.
If it was just "that easy" then I doubt only "Chinese models" would be doing it and we'd already be packed with competition.
Distilling might be a thing but it isn't a free win.
That's not the point. Why is it a country thing? There are plenty of non-China startups in this space having resources at that scale. The "China" has resources is some "Western media narrative" speak. So Meta should have won a long time ago? Or xAI?
> culture (Asians are generally collectively-inclined, so sharing is in their core)
Just stereotype it? So we've gone from China -> "Asian"? Then where is your Korean or Japanese model etc? And somehow you know they're sharing.
> political bent (there will be no diplomatic repercussions) to put up a fight
More inferring from "Western media news"?
Where's the reality?
The media hyped up Gemini / Google TPU free-win last year. How did that go?
Because the China vs US geopolitical situation is a thing. Meta is a social media company, not an AI company, and they direct their focus as such. xAI just never got serious traction so now they're selling their compute. Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
> Just stereotype it?
Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
The "and" that joined those 3 items is very important: it means you can't pull them apart and address them independently as they each contribute to the context. I'm not too sure about Korea, but in a way Japan is a US colony in all but name. Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens.
The reality is that China and the US are essentially in a trade war, where the latter is trying its best to keep the former in the Dark Ages, because "national security", but the former is refusing to take it lying down and continues to make progress regardless[0], because they have the resources and will.
[0] https://thenextweb.com/news/china-lineshine-supercomputer-to...
By the media? It's easy to point fingers at a blackhole.
> Meta is a social media company, not an AI company,
Alibaba (the discussion here) is not an AI company too (by your definition).
> Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
Meta has been to congress. Microsoft, Google etc have been in lots of court cases and continue to do so. Do you really think that is what stops them?
> Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
This is exactly the "media" view you get. It's just stereotypes and generalization.
And yes, that is wrong by the way. Evident in real data. "China" as a whole wins market share in many areas but no 1 company has as much of a monopoly as US companies do. Why? There's so much competition that it is scary. So are you sure they don't compete?
> but in a way Japan is a US colony in all but name
Again, I almost give up seeing this. Clearly, not. If a whole country, the world's top 5 in GDP is only that to you something is wrong with what you're seeing - not with the country.
> Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens
On the table? You do know that China is a top trading partner with all of these on your list. Despite whatever spat you might see in the media.
> The reality is that China and the US are essentially in a trade war
No. That's what the US government wants you to believe. It was even documented that in his 1st term, Trump, wanting a grand policy asked Krushner, whom then suggested China (pretty randomly) and so they went with it. Trump has now done less "China" related things lately due to all the backlash that you'd think he has moved on and found new toys.
Until very recently, the export ban GPUs had such a loophole that Chinese companies were able to use subsidiaries outside of China to buy and train that the whole thing was meaningless.
i.e. conclusion: stop getting brainwashed by media articles. It's all a show to get someone like you riled up.
So that was the real reason for the Fable restriction? Because Anthropic wrote a letter to the US government saying that China was distilling Fable?
Gosh, overusing accounts running up unplanned-for expenses?
Kinda reminds me of...overusage charges and inflated expenses clients have had to deal with because Anthropic, OpenAI, Grok, etc have been "illicitly extracting" everything they can grab from said websites, as fast as they can. In what amounts to a DDOS, frankly.
- Entitled jerk that initially wronged people
/Anthropic-probably