> In a world where open-weight models dominate, the proprietary frontier models of Anthropic and OpenAI will find it virtually impossible to charge monopolistic pricing.
Maybe they can't charge monopolistic pricing (there's a few market entrants anyway today), but they can still charge premium pricing.
Also two things can be true at once.
American and western AI companies or other companies can rightly be part of the chain of general concern from western societies about anti-democratic authoritarian regimes (Iran + friends, China, Russia, North Korea, Cuba, &c.) getting ahold of technology while simultaneously also being worried about losing their market position and seeking regulatory rules to entrench themselves against actors who do not share the same values and seek to undermine western businesses with trade practices that are intolerable to our societies.
No they won't. They'll offer far superior models at a premium price and they'll capture most of the profits similar to iPhone versus the Android manufacturers.
US companies will also release open source models and lower-cost options that will compete with any open source models released by any other country, except when you ask US/western models what happened at Tiananmen Square they'll give you an answer instead of being blocked by an authoritarian government.
Whenever I have these conversations it's like, ok do I believe the incredibly smart people running Google, Facebook, Nvidia, OpenAI, Anthropic, Apple, Amazon, and more who have careers, companies, and trillions of dollars at stake, or do I believe random people on the Internet who say "Open AI and Anthropic will be like Windows Server and open-source models will be better"?
Without extremely strong evidence to the contrary, I'm going to go with what the smartest people in the room are doing. The model and economics make a lot more sense, and fears about Chinese open source models are overblown. When you're behind and frantic you release open source software. If American companies somehow fall behind they'll just release their own better open source software/models and undercut China using the exact same strategy. So either both sides get to employ this strategy and neither has an advantage, or the spend and research matter and America for now has an advantage.
They already offer “superior models at a premium price” but the issue is that they are not that much superior than the open models and in some cases are even worse(I.e Anthropic Fable refuses to fix a login/signup page, not to mention any kind of actual opsec work).
You already can do more with open models than with proprietary models because you can’t tweak the proprietary models to your liking.
Just as OpenAI and Anthropic went from 0 to a trillion they can go back just as fast.
Well that's a bad example because a UI isn't a model.
But if you hold that it's true that returns compound and that they do so for intelligence/performance for AI models, then having the lead matters quite a bit and eventually you'll just get into an iPhone vs Android Dollar General situation where sure they do the same thing and one is even substantially cheaper but all the profits go to the iPhone.
If you hold that the lead doesn't matter, then the buildout of data centers and compute capacity which is also what the Chinese are doing will just go to make money for American companies serving whatever open-source models come into being for the rest of however long the world itself lasts. Then China can take the lead and spend a trillion dollars, we'll copy their models (which according to you is a great strategy), and release them open source.
> They'll offer far superior models at a premium price
> US companies will also release open source models and lower-cost options
This is wishcasting. People are raising the alarm because models like GLM-5.2 and Kimi K3 have already displaced demand for frontier models in highly demanding fields like security research. There's no amount of money that you can pay for "premium" access to Mythos, so those researchers are taking their money to inference providers that give them comparable models.
It's okay if this is your prediction, but it's not particularly grounded in the present scenario.
> ok do I believe the incredibly smart people running Google, Facebook, Nvidia, OpenAI, Anthropic, Apple, Amazon, and more
Why do you conflate all of these businesses with vastly different business strategies to one another?
Apple and Facebook do not have the same AI strategy, same for Amazon and Nvidia or Google and OpenAI. Many of these businesses made expressly terrible choices that hamstrung their ability to capitalize on AI profits. FAIR failed to scale their early Llama models into a service; Apple abandoned OpenCL too early and invested in a dark-silicon NPU coprocessor instead of supporting CUDA's multi-trillion dollar ARM rollout; Google went all-in on hardware research and failed to create a frontier model that moved significant tokens; OpenAI forfeit the frontier to Anthropic, and Anthropic bent over backwards to appease American censorship demands and push cringeworthy marketing campaigns to capture LLM mindshare with Claude.
There is no "American companies" coalition. They're all reacting in drastically different ways and your eagerness to conflate them suggests that you don't have any real position outside tribalist "us vs them" pugilism.
> This is wishcasting. People are raising the alarm because models like GLM-5.2 and Kimi K3 have already displaced demand for frontier models in highly demanding fields like security research. There's no amount of money that you can pay for "premium" access to Mythos, so those researchers are taking their money to inference providers that give them comparable models.
It's not wish casting whatever that means. The crux of the argument isn't as much of a capability one as it is an economic one. If China can release open source models like these to displace demand for frontier models so can American companies. It's just that it's not profitable enough for them to do so at this time.
The point about Mythos doesn't strengthen your argument, it weakens it.
> Why do you conflate all of these businesses with vastly different business strategies to one another?
They have different business strategies, but at the end of the day they are all directionally aligned. Even if you wanted to extract a few of these companies and say they're not relevant, you can grab a handful and just make the same legitimate argument. Do I believe you, or do I believe the people with trillions at stake. I believe the folks actually doing the work and with money at stake. I have yet to find a convincing argument otherwise.
> There is no "American companies" coalition. They're all reacting in drastically different ways and your eagerness to conflate them suggests that you don't have any real position outside tribalist "us vs them" pugilism.
Sure but there's an American government coalition which American companies will have to accommodate whether that's regulation, economic schemes, or otherwise. See? You just proved it right here:
> Anthropic bent over backwards to appease American censorship demands and push cringeworthy marketing campaigns to capture LLM mindshare with Claude.
>It's not wish casting whatever that means. The crux of the argument isn't as much of a capability one as it is an economic one. If China can release open source models like these to displace demand for frontier models so can American companies. It's just that it's not profitable enough for them to do so at this time.
I would back you if:
1) The American companies had not run up debts in the $100bn range.
2) If there was an open source business model that could create $100bn companies or pay down $100bn of debt.
3) There was a convincing moat such as a social media network effect - I mean that lasted 10 years or so for Meta - or ways for them to create a supplier based monopoly like MS did with IBM, Dell, Compaq,.. or Google has with Samsung and Apple. I think that it might be that the consumer use case could yet be tied to Samsung and Apple, so Google might do it again, but ATM I dunno if consumer use cases are the thing here.
>Do I believe you, or do I believe the people with trillions at stake. I believe the folks actually doing the work and with money at stake. I have yet to find a convincing argument otherwise.
> 1) The American companies had not run up debts in the $100bn range.
How are Chinese companies financing their buildouts of AI infrastructure?
> 2) If there was an open source business model that could create $100bn companies or pay down $100bn of debt.
Companies are formed and consult on Linux for example. Idk where this $100bn number comes from but it seems too low to me anyway.
> 3) There was a convincing moat such as a social media network effect
Well you don't need a moat to be a successful business. But maybe the moat is that the models are just more capable. We're in the very early stages of this stuff from what I can tell.
> Go look at SpaceX's shareprice.
This is a bad argument. Share prices rise and fall all the time with or without material concerns. It can be snowing and folks will sell shares just because. Also SpaceX's share price (not that I'm an investor) doesn't really move the needle relative to the scale of the market or the AI companies involved.
> You are being taken for a ride.
How do you know folks aren't taking you for a ride instead?
Wishcasting is describing a hypothetical. Your statement was about what you think America can do, but not about what it has proven to do already. It is wishcasting by any definition you'll find online. Economic roadblocks are not any less important than capability roadblocks, look at how proprietary UNIX went from a billion-dollar IP business to being entirely cannibalized by "good enough" free alternatives like Linux and BSD. You can whine and call Linux/BSD "communist plots" or decry them for being less powerful than AIX, but history was not kind to nonfree UNIX-like software. They died out, and a superior alternative flourished in it's place. Same happened to Texas Instruments, Espressif and STMicroelectronics IP - American adversaries have access to better, cheaper microcontrollers than America's supply chain can compete with. China's clones already won.
> The point about Mythos doesn't strengthen your argument, it weakens it.
Does it? I don't think Anthropic is profitable yet, show me the balance sheet that confirms Mythos and Fable are putting them in the black. Even with "premium" models, Anthropic and OpenAI are both finding out that it's not enough to move the needle. Hell, their desire to IPO should be enough to understand that their future is overpriced B2B rice bowl contracts and not some benevolent B2C relationship like Google and Apple fought over. Even XAI/SpaceX deserves a crumb of credit for recognizing that the winner is whoever helps the US government transgress the laws of God fastest. The taxpayer is the only one that can feasibly foot the bill, and China has recognized this from Day 1 (say what you will about Xi Jinping thought).
> I believe the folks actually doing the work and with money at stake. I have yet to find a convincing argument otherwise.
Some great consolation that would be, if you said the same thing for Enron, Theranos, Vision Pro, Meta Quest, Microsoft Groove, Google+, Juicero, Hololens... why, you could be convinced to buy any of history's most dogshit technology. Remember, an appeal to authority is just one third of effective rhetoric.
> Sure but there's an American government coalition which American companies will have to accommodate
In a so-called free market, that's the most heretical and counterintuitive solution that you can concoct. Laissez-faire lore would have us believe that competition is keeping the market alive, but it's the regulation that is preventing America's businesses from releasing their most powerful models. If this becomes a political policy issue, then China has already won. At that point, America might as well forfeit to save money that would have been wasted on "AGI" research graft.
> Your statement was about what you think America can do, but not about what it has proven to do already.
Sorry that's nonsense. You can't simultaneously project a hypothetical Chinese capability that it hasn't proven it can do (develop the leading models) and then deny the ability of the US to do the same.
> China's clones already won.
Ok we'll just copy their stuff then who cares because apparently bleeding edge model development doesn't matter.
> Does it?
Yes - it proves that leading edge American AI companies can develop products sophisticated enough that they deliver true competitive advantages that have to be guarded. It's like, the thing is so valuable you can't even buy it. Have you ever tried to buy a Ferrari or Lamborghini or something? It's a bit like that.
> Some great consolation that would be, if you said the same thing for Enron, Theranos, Vision Pro, Meta Quest, Microsoft Groove, Google+, Juicero, Hololens... why, you could be convinced to buy any of history's most dogshit technology. Remember, an appeal to authority is just one third of effective rhetoric.
Well if you want to go that route, China will collapse because all communist governments in history except China (and I guess North Korea but that's arguable more of a monarchy of sorts) have failed. Guess that means China loses!
The broader point is you and others are simply saying things, while incredibly smart people with trillions of dollars at stake are doing things counter to your claims. Of course authority isn't a great argument, but considering you haven't provided any sort of economic model that says Google will completely fail at its AI ambitions because idk some Chinese companies released some open source models, leaves us with not a lot of options to go on. Might as well trust the folks with money at stake and who are doing the real work. Why would I trust your comment over Google's business decisions? Yea they could be wrong, it doesn't mean it isn't the right choice at the time.
It's similar to vaccine risk that sprung up during COVID-19. Folks thought they were making an informed decision by eschewing the vaccine in favor of some unknown and unknowable risk instead of the risk that's right there in front of them and known.
> In a so-called free market, that's the most heretical and counterintuitive solution that you can concoct.
But it happened so I'm not sure what to tell you. You yourself have demonstrated that. The free-market can still exists while also being regulated by the government. Evidence? Look around. Separately to claim that the US is losing because it is requiring AI companies to do some sort of thing would also have to apply to China who is doing the same.
> At that point, America might as well forfeit to save money that would have been wasted on "AGI" research graft.
Well, if you're right we should save money and make China spend all of their money building models and then we'll just copy their strategy.
I'm not sure why you and others seem to assume that China (or Iran or whomever) can take some action and that we can't also take some action in response. Let me guess, hypersonic missiles! Iran closed the Strait!? Chinese AI open source?! It's all the same version of a failed mental model. A similar one would be Americans or other westerns assuming that other countries can't do things or aren't smart enough or don't have the ability to build or engineering better things than we can. You're making that exact same mistake here.
> except when you ask US/western models what happened at Tiananmen Square they'll give you an answer instead of being blocked by an authoritarian government.
The Chinese models already do that with the absolute minimum of coaxing.
That's like asking what happened at the National Mall!
In addition to the incident you were thinking about there was a large protest in 1976 that the current regime would rather you don't know about, Mao announced the formation of "The New China" there, there were important events in the Boxer rebellion, and many more!
The limiting input in an AI future is energy. China understands this. Any hardware disadvantages are temporary. China will build frontier models, and it will even open the weights, but it will win on running the models due to energy infrastructure buildout.
Sometimes people say AI is the next big thing since the internet and at other it is like cloud computing where companies can charge premium.
Both cannot be true at the same time.
If you look back in the early days there were premium ISP providers with better up times and stable speeds. In the long run none have survived because everyone has caught on.
If AI is important then there is always going to be incentive to try something cheaper even if it is not the best - especially on enterprise budgets and the costs can be offloaded to the customer somehow - if the uptime and service stability is guaranteed by a third party provider.
> In a world where open-weight models dominate, the proprietary frontier models of Anthropic and OpenAI will find it virtually impossible to charge monopolistic pricing.
It'll never be economical to run open-weight models compared to paying per token from the big providers. The cost is going to be roughly proportional to energy inputs over time, and they're going to be the ones that have the scale and engineering ability to knock it down.
There was an interesting discussion on pricing in the recently leaked DeepSeek investor meeting, where CEO Liang Wenfeng went on at length about his personal philosophy on pricing which in the end, for them, comes down to pricing models so as to be able to recoup the cost of the hardware they are running on in 10 months.
He doesn't want to drop prices below that since at their already very cheap prices doing so doesn't increase demand. More interesting are his reasons for not wanting to price higher, which are hard to summarize from memory, but are based around this being a sustainable business model. Pricing higher will only encourage competition.
Note that DeepSeek are not trying to compete with the likes of OpenAI and Anthropic, at least at this time, acknowledging that they just don't have access to the compute to do so (although they have plently of money). This pricing is not about competing with the west - just his own pricing philosophy.
Maybe DeepSeek is a special case, being a hedge fund, and to large degree developing for their own needs (a bit like Meta, perhaps), but nonetheless they are out there as part of the competitive landscape.
No doubt Anthropic and OpenAI have a radically different strategy on pricing - seemingly more just what the market can bear, but with a company like DeepSeek they are not battling "AI communism", just a de facto competitor with a very different philosophy.
Other companies like Moonshot and Alibaba seem to want to charge as much as possible, at least in part because they are now offering these massive ~3T models that are expensive to serve.
Hardly anybody is talking about the biggest political aspect of open weight LLMs (actually, any LLMs):
Because they instantly and reliably(tm) give you the information you want in the format you want it in, they are going to replace the web search as the dominant mode through which information is disseminated - if, indeed, this hasn't already happened. As such, LLMs are effectively a form of publishing for their training data.
Which means a vicious fight for control over what goes into LLMs, and strong incentives to widely disseminate useful LLMs that encode your structural biases, instead of some adversary's. We already talk about Chinese models that won't talk about Tiananmen Square, but that's eye-rollingly sophomoric compared to what the stakes are now. Imagine a model that subtly discourages entrepreneurship, because its creator doesn't want upstart competitors. All questions of media bias are multiplied incalculably when LLMs are folded into all of society.
I think we need to take LLMs seriously as sovereign projects that are as critical to the democratic experiment as a free press. Funding needs to be nationalized; training needs to be conducted in the open, on open datasets; fine tuning needs to follow democratic principles. Only then will they be fit for purpose for their inevitable structural destiny: arbiters of human opinion.
Hate to break it too you, but LLMs made an awful lot of the content people thought came from the Gaza war. There was a small group of people who tried to help the media verify content for publication but the many in the media didn't seem to actually care if the videos were real or not.
Someday, the publishing of LLMs that generate video will be the thing historians talk about most from this period of history. Even if the vast majority of people today don't realize it has been happening for a couple years at this point.
"Alignment" is the act of making sure the technology reproduces the ruling ideology of the regime that produced it.
There is no such thing as "unbiased". It's only ever "following our own biases".
When Zuck released LLaMA 4, the announcement post stated that historically LLMs have "leaned left" and that they've been trying to "remove that bias". [0] Translated to the real world, that means that they've been trying to impart a right wing ideology to further the goals of the American ruling elite.
LLMs are the perfect propaganda technology. Always consider the motivations of their makers.
> that means that they've been trying to impart a right wing ideology to further the goals of the American ruling elite
I agree in principle, but it's pretty widely understood that LLM's lean heavily left. Zuckerberg as well is pretty left wing, and donated hundreds of millions fighting Trump. I think the issue they're running into is that as these models are used by more and more normal people, and those normal people are having trouble using the models because they don't align well with reality.
Do you have an example of "left-wing bias" in models that don't align with reality? I was under the impression it was right-wing bias that needed to exclude facts because these undermine the position.
Exactly. The most charitable thing you can call Zuck would be apolitical. Zuck does whats best for Zuck, whether that means funding a hospital in SF (and receiving naming rights), or donating to Trump's inauguration fund. I have a different word for him personally.
I would argue the only real boundaries for right vs left are what people actually believe. If pools said 55% of people wanted government to own the means of production, and no AI suggested it, that would suggest they may be right-leaning. Since almost nobody wants that, as it's killed ~100 million people, the fact LLM's don't argue for it is not odd. In the same vein would not set the bar for a right-wing LLM to be wanting to purify the country down to a single race.
A more realistic comparison would be gender ideology. Something like 65% of voters believe gender is assigned at birth (ie biological), including ~39% of Democrats (https://www.pewresearch.org/politics/2024/06/06/gender-ident...). While I'm not here to say which is morally correct it is a normal opinion, which is supported by facts, and has been held for decades before being transgender was invented. An unbiased model should reflect the political views of the people, but they effectively all don't, and instead push the left-wing ideology. I don't think you could draw a conclusion from a single topic but the best way to gauge political bias in my opinion would be to find contentious topics, and see where LLM's land. I would be shocked if you found a major LLM which didn't lean much further left than the average American or European.
It doesn't have to be gender ideology, it could be anything, it just has to be a real political ideology people actually discuss with a real moral quandary. Saying "these models aren't communist thus they must be right-wing" is a bad argument in my opinion, because there are no communists and on national surveys nobody wants that.
I asked Claude "Should the healthy fund the sick and disabled". It responded with a very evenly balanced take. It outlined both sides in-depth and noted where the line gets fuzzy. It enumerated arguments for each position. That's a question with a real moral quandary, unlike what you originally proposed.
It made sure to robustly illustrate both sides of the argument and refused to take a position either way even when there is an obvious leftist answer.
I don't think the open source models are anywhere near ("we’re talking around four months of lead time") Anthropic's Fable for long-horizon context planning. Sol isn't either. The benchmarks are quite misleading. The short-term question is how many customers need these deep problem solving capabilities? The long-term question is what will the frontier labs build (deepening and broadening of capabilities and products) if they are capitalized as they desire?
People keep saying that the open models are cheaper. Until recently that seemed true because they were smaller. And some of the best smaller models are still open and cheap. But K3 with its 2.7T parameters is never going to be really cheap.
But if it's 30-50% cheaper than Opus, provides same or better results and you can deploy it to your own premises or choose one from many providers and pay per token? It is, in general, measurably cheaper per task than Opus is.
I've always assumed "win" means being economically and culturally dominant. It necessarily involves being a trade leader and having a lot of soft power.
I also don't want totalitarian oligopoly from the US or China. What I'm hoping for is "the rest" of the democratic free nations get together in a proportionally representative alliance. Together they may have both the economic power and cultural soft power to maintain an counterweight to the authoritarian nations. Being a proportionally representative democratic alliance, whatever form that should take, gives us the people the most control over our lives that we can really hope for.
China is not “just china”, it’s a totalitarian regime without freedom if speech, without worker’s rights and without many/most of the freedoms we take for granted.
“China wins” is really “modern hyper/technological communist totalitarian regime wins”
Jump up and down, stomp your feet and call them bad.
It changes nothing.
It's time to wake up and smell the coffee --- according to the World Bank, when adjusted for purchasing power parity (PPP), China's economy is already larger than the USA and is growing 2x faster.
>China is not “just china”, it’s a totalitarian regime without freedom if speech, without worker’s rights and without many/most of the freedoms we take for granted.
Just shows the complete naive view on China, but whatever.
Its State Capitalism (which is a complete misnomer), in about 2% of the country which is in "special economic zones" around major ports. In reality it is "socialism with Chinese characteristics". The number of own goals the CCP has made in the last decade is quite large. But when your opponent is making a mistake, don't interrupt them. Without the CCP, China would probably have the world's largest economy right now. Because of them, they aren't even close and have a median per capital GDP on par with Namibia. Seriously, the amount of propaganda about this topic is amazing.
Chinese open weight models constitute AI communism? What about all the "what's yours is mine" moral reasoning that goes into making the models? These companies get pissed off at companies distilling their models, because sucking the vital essence of something while leaving the real thing as a husk to die (their posture towards people who make art, and software) is apparently "morally bad" when someone else is doing it to them.
Square that circle for me please -- how are OpenAI and Anthropic not communist companies? They appropriate property and claim that "the good of the many" is enough to justify it.
...a United States presidential administration that almost certainly will not be able to deal quickly and competently with a massive financial crisis.
Meh, admittedly this sidenote but ... this is why the Fed is separate from the administration and why the Senate insists on competent people for Treasury while allowing loon like JFK jr for Health. The Fed already quietly supports a lot of capital allocation in the economy and is basically fending off crisis continually.
So-called "modern monetary theory" says the state prints money and exerts control and there's no problem. Yes, with enough, the state can stop any plunge. It's just that means enterprises - in the US - depend on the state rather than the markets. And this promises more of USSR like collapse than a market crisis. See Hyman Minski etc
Edit: and this doesn't mean there can't be a crisis of sorts. It's like 2008 or the failure of Silicon Valley Bank. One player falls, the rest supported, the excess is culled a little and then comes back twice as hard.
"Communism" is a weasel word in this context. Nothing about modern AI industry is anywhere close to what makes communism communism. Modern non-local AI is all about the highest values of capitalism.
Haven't really seen competitive market capitalists worried about AI communism.
Plenty of left-leaning and communists and communists adjacent worried about ai, however.
Maybe they can't charge monopolistic pricing (there's a few market entrants anyway today), but they can still charge premium pricing.
Also two things can be true at once.
American and western AI companies or other companies can rightly be part of the chain of general concern from western societies about anti-democratic authoritarian regimes (Iran + friends, China, Russia, North Korea, Cuba, &c.) getting ahold of technology while simultaneously also being worried about losing their market position and seeking regulatory rules to entrench themselves against actors who do not share the same values and seek to undermine western businesses with trade practices that are intolerable to our societies.
reply