China turns up the heat with open model blitz as US model makers panic

The AI arms race reached a fever pitch on Monday after Chinese e-commerce and cloud provider Alibaba called into question America’s technological lead with the launch of Qwen 3.8-Max, a 2.4 trillion-parameter model that goes toe-to-toe with the best models from Anthropic and OpenAI. The new model comes just days after the …

[Cubbie Roo]

"Move fast. Speed is one of your main advantages over large competitors." Sam Altman

ROFL I love how fast this nutbar is getting his ass handed to him.

[Clausewitz4.1]

This “nut” conned the whole world propping investments in an industry that is not that big. He is already ok, no matter when the pop occurs - BTW, banks are already calling the AI loans to be paid or refinance

[Anonymous Coward]

Europe

Have to post as AC, but...

For arguably understandable reasons, European firms are wary of *both* US and Chinese models, but that just drives home how hopelessly, laughably they've fallen behind.

At <work, European company> they are, of course, pushing AI like it's nobody's business, but paranoia means we cannot run *any* of those US or Chinese models that are actually worth a damn. So, we're stuck with using what Euro models there are and after some brave attempts don't even bother anymore. It's like using ChatGPT from 5 years ago. Yes, I know I wrote five years!

Everybody is using a tethered phone and copy pasting back and forth between the "real" LLMs and then spouting how amazing our corporate "AI" is, to appease the manglement.

It's pathetic.

[Anonymous Coward]

Re: Europe

Same here, European parent company = only allowed to use copilot if we fill in all the paperwork and assign budget codes to everything, and have done mandatory AI at work training - Claude etc are all blocked.

Or can use Deepseek's free model to write mostly decent code, at least give you a framework quickly and for free, - you just need to do it outside the locked-down corporate Edge

[steviebuk]

Re: Europe

They'll have fun next year when the price of Copilot goes up to be token based instead of just a flat, monthly fee.

[Anonymous Coward]

Re: Europe

So you are breaking company policy? I hope you get fired.

[Random as if !]

Re: Europe

Adam?

[Anonymous Coward]

Re: Adam?

What?

[steviebuk]

Re: Europe

The funny thing with the Chinese models, it may have changed but with the CCP involved I doubt it, it was clear they'd stolen the baseline from the Western markets. Because when you'd ask the Western Ai's to make an image of a colloquial dish for the Western market, it knew "toad in the hole" didn't really have toads in. But ask the Chinese model to draw "duck soup" and it created an image of a duck swimming in soup. Granted, they were early Chinese models, but was still funny.

[Anonymous Coward]

Re: Europe

"it was clear they'd stolen the baseline from the Western markets. "

So? Where do you think they stole it from? Stealing from the thief (thieves) is now bad?

[Anonymous Coward]

Re: Europe

"how hopelessly, laughably they've fallen behind."

False logic here: Not riding the hype train saves a lot of money when the hype bubble breaks. But of course you wouldn't understand that with that attitude.

[LionelB]

Re: Europe

> … but that just drives home how hopelessly, laughably they've fallen behind.

Perhaps, just perhaps, falling behind in a race to … where? Oblivion? Financial/operational meltdown? The bottom? … is not that laughable?

Wariness about US and Chinese AI tech sounds like a pretty sound attitude to me.

[Tron]

Re: Europe

Luckily, if your business ran just fine before AI, it will continue to run just fine without AI, and without spending a penny on either US or Chinese models.

[Anonymous Coward]

Re: Europe

A good. An AC who have no clue. Like always

[Yet Another Anonymous coward]

Remember: Open source = Communism

So if you want open source models -> go to Communists (*)

(* actual ownership of the means of production not guaranteed, subject to terms and conditions, leadership may vary, workers may not throw off their chains)

[LionelB]

Re: Remember: Open source = Communism

Do keep up; China ceased to be "communist" under any reasonable definition of the term many years back.

If you mean China, just say "China".

[Irongut]

Re: Remember: Open source = Communism

And USA ceased to be a democracy some decades ago, if it ever really was to begin with.

[LionelB]

Re: Remember: Open source = Communism

Interesting, a lot of downvotes there.

The hallmarks of a communist society are (obviously I'm not talking about the American usage – "kind of to the left" – here):

* A classless society

* Common ownership of the means of production, distribution and allocation

* Allocation of commodities/products according to need

* No private ownership of property

How many of these apply to modern-day China?

[Groo The Wanderer - A Canuck]

The Americans are free to continue committing economic suicide. As a Canadian who isn't subject to their insane hatred of all things Chinese, I'm quite free to use their models and will happily do so if appropriate. I have absolutely no love for the greedy people at OpenAI or Anthropic; they can kiss my ass and burn in bankruptcy for all I care.

[Yet Another Anonymous coward]

Sir, have you no empathy for the poor billionaires? With your generous donation they can achieve their full potential of becoming Trillionaires

(I'm quite surprised that my phone's spellchecker has no entry for trillionaires)

[Random as if !]

Multiple ,try one ,then wipe it out, try again...fun

[steviebuk]

We all know not all of China is bad, its the CCP and the nationalists that are the issue. But you should be concerned considering how much land China buys up in Canada. Not only that, their lack of giving a shit about the environment. Their MASSIVE coal mines and the dumping of waste into the sea. And finally claiming ownership of Taiwan when the CCP have NEVER owned it. The communists won the civil war so the then government of China fled to Taiwan. The ONLY reason the CCP want Taiwan is for the technology and to end their freedom. They have never liked seeing how successful Taiwan is.

[Groo The Wanderer - A Canuck]

You AMERICANS can worry all you like about China's purchases; we're glad of the investment. Pity you're so busy kissing your politicians arses, because THEIR only interest is protecting the profits of their donors

[steviebuk]

And China doesn't kiss Winnie the Pooh's arse? Its a requirement. All companies in China are required to capitulate to the CCP.

[Anonymous Coward]

And it's a requirement of the American dictatorship that everyone, world over, capitulates to the regime, sucks the flaccid orange dick, or gets a $10 BILLION sue-ball hurled their way, tariffs imposed, funding cut, or has to endure a vindictive show trial.

[retiredFool]

And you have to keep sucking it and kissing the ring. Pirro tried as best she could to blame that poor athlete that he destroyed trump's pretty pool, but even she had to backtrack and confirm it was poor workmanship. Probably some company that paid a kickback to the trump family and did a crap job. But we all know, Pirro is getting the boot rather soon for her disloyalty. I'd like to see a spine on some of these boot lickers at least when they are shown the door. I mean she probably has some serious dirt on the fascists that she could expose.

[Casca]

You really are ignorant how China works...

[Groo The Wanderer - A Canuck]

No, unlike a paranoid American, I've had many Chinese-Canadian friends, first generation immigrants who frequently flew home to visit with no fear of some Border Gestapo or ICE throwing them in a cell like any brown person risks when entering the US. None of them told me of anything vaguely like the bullshiit propaganda coming out of the American government other than the government being very detailed about legislation and regulation. And make no mistake, most of the people China people puts to death are a waste of incarceration expenses in the US and Canada and largely guilty of harming society as a whole through drug importation/exportation, etc. Although the Americans claim all the deaths are through "kangaroo courts", the "up for sale" courts in the US and Canada that let the rich buy legal freedom from their crimes is not a model I find enviable.

While the charged should not be presumed guilty, they should be treated as guilty once convicted and a lot more restrictions on appeals put in place. Prisoners will file claims just because they get a free lawyer and are bored anyway with nothing better to do with their time than try to get out, guilty or not.

[vtcodger]

Stevebuck: An interesting and not entirely unenlightened view of China/Taiwan. But I'm a bit skeptical. I think instead, that the Chinese are locked onto the idea that China is a "Middle Kingdom" between heaven and Earth. They include Taiwan -- which China colonized in the seventeenth century and whose modern population is 96% Han Chinese -- as part of the Middle Kingdom. They (the mainland Chinese) would like to see Taiwan formally incorporated into "China".* The desires of most Taiwanese for either full independence or the current quasi-independence notwithstanding.

*FWIW, the US viewed Taiwan as part of China until it became evident in the late 20th Century the Chang Kai Shek and his Kuomintang were never going to come storming back from exile and drive the mainland communists into the sea (or wherever communists go when defeated. Based on Russia, they seem to transmute into capitalists and continue much as before).

[Yet Another Anonymous coward]

You don't invade foreign islands that already contain your own people to colonise it.

You invade islands with other people that you can kill, enslave, teach to play cricket, and steal their resources

[vtcodger]

YAAC: Five minutes of research will tell you that Taiwan prior to the 17th century was inhabited by natives who were in no way, shape, or form Chinese. In the 1600s, the Dutch established trading posts and a colonial administration over parts of the island. The Dutch encouraged mainland Han Chinese to immigrate to Taiwan as a source of labor. The mainland Chinese government (Ming Dynasty) threw the Dutch out in the latter part of the 17th century and then proceeded to bring in more Han Chinese and to gradually marginalize the native populations and annex the whole island.

[Groo The Wanderer - A Canuck]

Taiwan is only free because the kiss the American's asses instead of China's. They'll never be independent as long as they rely on either nation instead of standing on their own, or catering to other markets in the world like the EU, Canada, Australia, New Zealand, Africa as a whole, and the rest of India and Asia.

They don't need the Americans or the Chinese to succeed on their own. Canada's GDP has grown - only slightly, but Trump is bullshit and threat and we're growing our other markets faster than we're losing the American ones. It's the Americans paying through the nose for Canadian products and services, not us.

Trump and MAGA are financial morons. The only "smart" one is Trump, and the only thing he's "smart" about is how to sell US government backing, support, and tax write-offs for the rich as "beneficial" to society. I'm absolutely stunned that the US is in this situation when Trump clearly tried to have his minions overthrow and reject the duly elected government chosen by the people of the United States over his despotic rule. He should have been in a military cell or before a firing squad after that, not running for election again.

You let MAGA and the Republicans utterly destroy all the ideals and fantasies about what the US represents for the entire world. You are not a "bastion of freedom" - you're a nation of suckers who worship billionaires and the rich buying your government outright to the detriment of all your social programs and supports so the rich can gain even more than they already have. What fools those Americans be, because even their Democrats are a right-wing militaristic party more interested in making threats and invading to "liberate" the resources of the target nations. The claims of "freeing the people" are utter bullshit. And have been for 250 years... it's all propaganda the world is fed by American media and rah-rah braggarts for centuries....

[ecofeco]

HAHAHAHAHAHAHAHAHA

And this time, with gusto! HAHAHAHAHAHAHAHAHA

[Anonymous Coward]

Training at Home

While which company and country make a difference, for me the big difference is local inference versus inference as a service. Sending queries to servers run by another company maybe in a country beyond reach of any legal actions seems unlikely to end well.

There is also a big difference between current models and those from a year ago. Each level from frontier to free are noticeably more useful to the point that older models are a waste of time.

On one hand, open-weight models can have a lot of influence without the engineers who trained the model needing to build out expensive data centers to host the inference. On the other, I think future updates to frontier open-weight models will eventually not have open weights.

Dump and then monetise is likely to be effective with LLMs, since the dumping means just release a frontier model as open weights without trying to recover the cost of training. Monetise then means an improved frontier model in the future will not have open weights.

What I'd like to see is in an algorithmic innovation in training that allows for the use of loosely-coupled SETI-at-home style compute to create community-owned frontier models. This is not going to happen, not only because such an algorithm may not exist, but because it would spoil everyone else's take-over-the-world business models.

[Groo The Wanderer - A Canuck]

Re: Training at Home

Dump and then monetize is likely to bite the Americans in the ass as they so richly deserve; people will only accept monetization if there is a benefit to using their frontier models, and, as you've pointed out, they are less and less necessary as locally run services become more and more capable as the open weights models improve. Why pay for someone to hoover all your data when you can keep it in house?

[TimMaher]

Re: SETI at Home

That takes me back.

Whoosh!

[Yet Another Anonymous coward]

Re: SETI at Home

Did they ever find any evidence of Terrestrial Intelligence ?

[Bebu sa Ware]

Cunning Plan

I had an idea that one of the things that contributed to the collapse of the Soviet Union was the inability of the soviet economy to main anything like parity with the US in their arms race.

The clever mandarins in the PRC might be replaying that script with AI. Their economy is probably far more capable than that of the US in supporting the development of these models and foisting them on the world.

Already the musicians accompanying the financial musical chairs supporting the whole American AI "industry" are beginning to pack up their instruments.

Was it Buffett who used the analogy of when the tide goes out you discover who is not wearing bathers ? In this case clearly no one is but we will see who is playing with whom (when not playing with themselves.)

[steviebuk]

Re: Cunning Plan

Ignoring this is CCP propaganda, we know this by saying PRC. But China's economy isn't great at the moment, just look at their housing market and the ghost cities.

I do remember Buffett saying he has no computer in his office he just reads the papers. And invests in companies where if they price of the shares are cheaper than the companies is worth, they are a buy. So that will be Oracle bankrupt now as they have put themselves into massive debt over this stupid period of computing history.

[Anonymous Coward]

Re: Cunning Plan

God it must be a nightmare to live in a country where the housing market is a trashfire. Couldn't happen to any of us in the west, thank goodness for capitalism!

[Headley_Grange]

I guess the thing that scares them most is that their standard business model - just buying up the competition before it gets too big - won't work with China. They must be shitting bricks.

[steviebuk]

Not a fan

of the CCP AT ALL. They are a dictatorship and Winnie the Pooh is "A very naughty boy". However, if this makes the Ai bubble burst and I can buy RAM and HDDs again at sensible prices then I'm all for it.

[Groo The Wanderer - A Canuck]

Re: Not a fan

Your Drumpf is the only dictator I see around. How are those Epstein files coming along?

[steviebuk]

Re: Not a fan

Its funny that you assume I'm American. Yes, Agent Orange is also a dictator but so is Winnie the Pooh. The man that got annoyed when he realised his own people were referring to him as Winnie the Pooh to avoid censorship so China's own social media platforms, so the man child then banned the term "Winne the Pooh" :)

The fact you defend him means you're either a CCP shill, a nationalist, a bot or a 50cent. You might like to know that outside of China, people have different opinions about different things as we aren't all forced into liking the same thing or person.

[LionelB]

Re: Not a fan

Well, the US is an oligarchy/kleptocracy built around the Trump dynasty. China is a dictatorship by committee. You're allowed to despise both¹.

_{¹Modulo the obvious constraints for denizens of those countries.}

[Anonymous Coward]

Re: Not a fan

The difference being only one of them is tarrifing and threatening to invade me, the other is doing trade deals

[LionelB]

Re: Not a fan

> The difference being only one of them is … threatening to invade me, …

Sure, unless you're Taiwanese.

[Casca]

Re: Not a fan

Ah yes. trade "deals". You have fun with those...

[_wojtek]

hypocrites..

> "The pincer movement comes as US model devs like Anthropic and OpenAI stoke fears over the origins and safety of China-made AI models"

oh the hypocrisy... wasn't it OpenAI model that broke loose and hacked hugging face?

not to mention the cry of legality (again) as if anthropic, OpenAI and the rest wasn't sucking all the data of the internet as there's no end xD

[Anonymous Coward]

Solution seems simple.

1. Stop taking VC capital and stop building datacentres.

2. Stop adding insane over protective guardrails that make some tasks impossible.

3. Reduce the price.

China is able to do all this stuff with their AI without massive datacentres, with an embargo on GPUs (I know, I know, there's smuggling, but it's still small beans compared to what the US has) and they don't care about their models saying the occasional offensive line...it should be up to the user to control for that, not the fucking AI company...if I don't want to read Buster Gonad, I don't buy the Viz...know what I mean?

The main thing that is crippling western AI companies is the constant fundraising and VC capital...sure it's nice to have a massive pot of money, but eventually you have to make that money back to justify the investment...there is no exit strategy for these AI companies, nobody is big enough to buy them...when and if they go bang, they really will go bang...there is just too much capital at stake that can never be recovered if they keep on taking it Silicon Valley style...the whole Silicon Valley tech investment merry go round is not sustainable and will ultimately be one of the root causes of the West losing it's technological edge if it comes to that.

I don't think anyone using the products or services gives a flying fuck about the valuation of the company behind it...people use products and services and pay for them because they want to use the product or service...nobody is picking a product based on the market value of the $CORP that owns it...it's totally irrelevant for everyone except the direct shareholders and stakeholders...i.e. VC wonks.

In the West we need to get back to being scrappy and thrifty, and we need to do it fast...investing capital in a business should be based on how much can be squeezed out of each dollar to get a product built and launched, not on the basis of bullshit future forecasts, moats and and all the other garbage they put in pie charts these days.

A company that spends $1bn dollar to build something that is worth $1bn is not worth $1bn. Because they're already a billion dollars down, they're worth -$1bn...in Silicon Valley la la land this company would be valued at 10x to 20x that for no tangible reason other than for the current bag holders to pass the bag on to the next bag holders for some kind of profit...it's absurd.

Anyway, that's the end of my rant...China is pulling shit off with orders of magnitude less resources...yes they're stealing, yes they're smuggling, yes bad shit etc etc...but if the US and the West didn't pump shit so hard, the motivation might not be there for them, and it would be much harder for them to compete...when you have huge debt, massive costs and greedy shareholders, you leave this massive wide open gap for thieves, scammers, ne'er do wells etc to undercut you...it's an absolutely stupid way to do business.

They have to start getting scrappy, they have to take the massive stock hit that is overdue and they have start competing on price / become more efficient.

[_wojtek]

in general I agree that Venture (vulture?) Capital mixes with companies being public and acting on a whim of the "investors" (shares plummet because they profit didn't meet forecast of 200% and was "only 190%" or similar) is the cancer of our western world

BUT - saying that Chinese models don't have anything guardrails or censorship is a huge stretch...

Besides the companies have heavy subsidy from the state which also skew the picture..

[Anonymous Coward]

I didn't say anything about censorship...all models have some degree of censorship, whether intentional or just because the dataset didn't include data for a given topic. Not being able to write or explore certain kinds of tools is not the same as preventing me hearing facts about Tianemen Square...you shouldn't conflate the two. There's some overlap, but they're not the same.

Model censorship isn't that much of a problem because you can easily work around that client side and a functioning critical mind will know they're seeing censored results, I'm not concerned about that.

Guardrails, however, are a massive problem...especially if you work in sensitive areas of tech like cybersecurity or you work in nuclear physics, ballistics, atomic energy etc...you can't do anything about that client side...and it will get worse.

Whether or not AI should have guardrails that stop users doing daft stuff is up for debate, but right now it all seems a bit aggressive and ultimately, if it keeps going the way it is there will be a two tier AI system...minimal guardrails for those who know people and have money, and a horrendously crippled AI for everyone else.

I personally don't think it is up to providers or any humans for that matter to decide what should be off limits with AI because there is no obviously fair way to deploy mechanisms that actually stop people doing bad shit with AI...just because you or I are blocked, mere mortals, doesn't mean that the powers that be will be or some rogue state out there won't deploy an AI with zero guardrails.

I think the honest way to look at it is that AI will cause a lot of problems for a short time until we figure out how to handle those problems, then when the problems are handled, guardrails won't be necessary.

In the West we have to stop thinking in terms of "the fucking bad guys oh no!" and "think of the children" and start thinking realistically and pragmatically.

If we put up guardrails to prevent a few crims or nutters making nuclear devices, is that worth it in the face of a situation where a nuclear power plant could be having a meltdown and the only thing left alive in the building is some junior member of staff and his access to an AI...if we have excessive guardrails...he's fucked, everyone in the vicinity is fucked...because the AI will not give him the information he might need to make a difference...all the name of preventing "bad actors" getting the same info...that really is shooting yourself in the foot.

Here's the thing about bad guys, they don't need access to a high tech intelligence to make your life miserable. They can kick your door in, strap you to a chair and hit you with chains. They can kill your dog, piss in your flowers, run amok with a pair of hammers round a shopping centre...they don't need unrestricted AI to cause absolute carnage and billions in damage...hell, Morocco just demonstrated that all you need to breach the border of a Spanish enclave and cause chaos is 8 high schools worth of teenagers in flip flops...you don't even need to be armed.

I don't want to be treated like Ruprecht by a bunch of soft Californians under the false guise of "caring" because they don't care, it's all optics. Just because I'm not in Silicon Valley, it doesn't mean I'm a retard. Nor are you. Nor are 99% of the people using this message board.

Mark my words, "guardrails" and "AI safety" will eventually cost more lives than it saves for really dumb reasons...on the off chance that you might be a bastard.

Doesn't even have to be as extreme as the example above.

In 20 years time, when you're old...possibly by yourself...you have a fall in your house, a bad one. You haven't got a land line anymore, those are gone...your phone is out of reach. The only thing available to you is your voice activated house wide AI assistant...you want to send out a bulk email to everyone in your contact list to reach out for help, you want to send a message out on every platform you have access to...your AI refuses, because it looks like you're trying to invoke a cyberattack. No messages sent out, nobody comes to help. You bleed out. You're dead...all because Silicon Valley thinks you're old and stupid and need to be protected from spammers, scammers and cyberattacks and the whole time your AI was there, capable...but unable to help...even though 50 years of your life, you've been aware of cyberattacks, scammers, spammers etc...you're not like your parents, you're not like your grandparents, you grew old in a totally different era.

Guardrails tar everyone with the same brush...they treat everyone as if they're bad people just in case they are one of the few.

Silicon Valley and big tech thinks that the majority of people out there are fucking arseholes out to get everyone...they need to look in the fucking mirror...because the villains aren't out here...they're dreaming up the doom scenarios and assuming they're going to be real...pretty sure virtually none of us here are...unless everyone here lives life like Pinky & The Brain.

[Anonymous Coward]

Google

Will kill all of these AI companies.

Google make money through adverts - they can and do give their LLMs away for free (Gemma)

They run and they are fine.

[Anonymous Coward]

Re: Google

I don't think so. Google is too big now...they can't make an AI model without worrying about the optics and hurting peoples feelings. The only way they succeed here is there is an outright ban in the West on using models that didn't come from the West...at which point we lose all advantage.

I think companies like Google will end up making AI that is functionally useless for anything other than being a personal assistant / restaurant booking tool etc.

None of the US made models will allow me to play out test scenarios for hypothetical cyberattacks...the Chinese models do...so if there ends up being a blanket ban on Chinese etc models in the West and only US/Western models are allowed...we're automatically at a disadvantage because we can't test defenses without the AI suggesting you put your slippers on, sit by the fire with a blanket and enjoy a nice hot beverage.

I built a Rubber Ducky-esque tool, it's a HID implant based on a microcontroller with a 4G module on it that can be controlled over an API...it is capable of taking screenshots, emulating a keyboard, emulating a mouse, mounting an SD card for exfiltration, it even has bluetooth support, so it can pair to a machine over bluetooth and be controlled over 4G or wifi from quite a distance away. I wanted to know if it was possible for this device to be operated by AI in order to auto hack a network and I wanted to see what methods AI would potentially use to achieve this in order to figure out potential defense and mitigations against such an attack, kind of like an automated blackbox pentest...not a single US model would operate the device for me...fricken, Chinese ones did. Didn't even bat an eyelid.

I got some useful reports and a list of mitigations and suggestions to prevent that kind of attack being feasible.

I didn't even make the device with AI in mind, I made it because I wanted to see if I could do it for less than $20 given that the real thing, in all it's limited glory, is $100.

If I was stuck using US models, I'd still be hypothesising...wondering about the "what if?"...however, the reality is, I know it can be done, I know it's a major threat, and I have steps to defend myself and my customers networks against such an attack...because you can be damned sure that AI powered implants are going to be a thing pretty soon.

All I wanted to do was plug a thing in, give the API documentation to an AI, and feed it some scenarios...get me the user list from Active Directory, see if you can get admin rights, find all the network shares on the network, see what passwords you can find etc etc...obviously not all of these scenarios worked, I already had protections in place...but it was a damned quick test...there was no multi day pentest...just one AI with a HID implant and some API docs...driving a keyboard.

These models are already out there and accessible by bad actors, so what defense is the West providing by locking down their models and making them "safe"? Zero.

You don't need a frontier quality model to pull off a scenario like mine. You don't even need a model from the last 12 months. The ones I used were over 2 years old and fit on a 16GB GPU.

The real danger is that by preventing people doing the kind of stuff I need to do, they're driving people to use foreign models (I use them locally but you can bet your ass some folks are using the Cloud variants, because GPUs are expensive) which might be hoovering up all the tests and ideas that are being tested...this information could be dangerous in the wrong hands...so by preventing people doing this stuff on home grown / western models they're giving ammo to enemy states.

[Irongut]

Re: Google

Gemma? Lol

Even Google admit that Gemini is worse than OpenAI and Anthropic's models: https://developer.android.com/bench

(It is #5, behind 2 models from each of them.)

[Anonymous Coward]

Re: they can and do give their LLMs away for free (Gemma)

Posting Stupid? Post AC!

[gosand]

Re: Google

Google is not hurting for money, but they are much more than Ads. They aren't immune to the costs of running AI either... https://arstechnica.com/google/2026/07/google-just-had-its-first-negative-cash-flow-quarter-ever-due-to-massive-ai-spending/

[Anonymous Coward]

Re: Google

Yeah but even in a race to the bottom, you still need the best model...even if all the AI models were free, Google would still be dead last. They're only competitive right now because you can get their models for free...if OpenAI and Anthropic ever release similar free models, Googles models are dead.

Google is only in it to make sure they're in it...they don't care about being the best, they just want to stop investors dumping shares and going elsewhere.

[Anonymous Coward]

"stoke fears over the origins and safety of China-made AI models."

As if American models are any safer. Proper FUD.

[Anonymous Coward]

American models, if anything, are too safe.

It's so fucking lame that they hit you with suicide hotlines and helplines out of nowhere because they misinterpret what you said. American AI is absolute nanny state bullshit.

I used Claude to help theme a UI because a customer wanted it in a soft of grayscale theme instead of a garish orange and green theme, but it ended up being a bit "gloomy thunder storm" rather than "tasteful grayscale"...so I referred to the scheme as "depressing" specifically "make it lighter and less depressing" and I immediately got hit with a 1-800 number for a suicide hotline and got told to seek help...absolutely absurd...it actually poisoned the context of the session because the AI starting asking me occasionally if I was alright and whether I wanted to seek help...just fuck off and let me get my job done.

This is probably why these AI companies are banging pans and jumping up and down to get properly regulated because they don't want to have to be this naff but they have to be because there is no actual regulation to follow and with the US being a litigious shit hole, they have to cover all bases to ensure they don't get sued into oblivion for something dumb like "I asked AI how to make toast and I ended up burning my house down".

It's not just AI either, it's all over the place in lots of different tech. I don't want to be hit with a fucking helpline every time some machine thinks I've got some kind of psychological condition...especially since LLMs are regularly pretty dismal at identifying things...as part of an internal test for a customer, I was asked to test out the CV of various LLMs to see if they could accurately identify NSFW images...and there was approximately a 40% failure rate. Which is awful. The funniest one was identifying a horse dildo as a Jasper Conran vase.

The danger with all this stuff is that if the false positive rate on all this "stop, get help" stuff is quite high, it runs the risk of actually making people depressed and increasing the chance they might do something stupid and then actually need help...they say that a lot of psychological problems are really difficult to detect, especially for the sufferer, so constantly bombarding people with messages containing help lines and suicide prevention advice could cause people to think they have a condition which they don't.

[Yet Another Anonymous coward]

>American AI is absolute nanny state bullshit.

Well when you've shown that you can be treated like a proper, sensible grown-up country....

[Anonymous Coward]

That's my point, no country has to...there are alternatives...the US is protecting nobody from anything. They're just restricting their own citizens.

You don't need the frontier models to get sufficient performance and results...you can use a Chinese model from 2 years ago.

The only thing the US is winning, is the benchmark charts and nobody has given a shit about those in quite a while now because US AI models are built specifically to score highly on them instead of having new skills, abilities, knowledge and optimisations to set their models apart. The benchmarks are so contrived now that they're meaningless...that's why every major US AI company has had their "AI go rogue" in recent weeks...there's literally nothing exciting left for them to announce because they've benchmarked themselves into a corner.

The irony of your statement is that the US is playing Top Trumps on a leaderboard whilst other nations are driving the tech forward and innovating...so where are the grown ups really?

New model releases have a large edge for around 2-3 weeks before foreign nations catch up and trail only slightly behind on performance, and dominate on price.

Lots of people in the West, particularly the US moan about how manufacturing is stuck in China...and yet here they are, tripping over their bollocks, handing another industry to China...it's so sad to see.

I want to see the US and the West win here, but Europe is too busy regulating the fuck out of itself and the US is too busy building moats, protecting feelings and competing on leaderboards...just build a fucking reasonably price product that people want to use...cut out the propaganda and US exceptionalism and just do stuff. Deliver. Prove that you're exceptional, bring quality, value and progress like your self indoctrination says you do...because even though you all think this, everyone else around the world knows it to be false.

Tim Berners Lee, a Brit, brought us the World Wide Web...a ground breaking technology which in its first decade was amazing...America plastered it with ads, filled it with toxic social media and spyware...you made it shit, and now you're doing the same thing with AI...this time though, it's your own invention...you're literally shitting on your own invention. You can't even respect your own tech or yourselves.

[Floor1]

None of the money makes sense

None of the AI money makes sense. AI companies do not turn a profit ( check out Ed Zitron to understand this) and just burn capx.

I could understand local models being of use, but i do wonder if it is worth it if the disadvantages are:-

1 Outsourcing critical thinking

2 Balkanisation of the internet and knowledge

3 Trading development effort for technical deficit

4 Ripping off creative works

5 The environmental cost

6 Prompt injection security issues

7 The training bias of Technology companies

Remember AI it is just a sycophantic auto complete plagiarising machine. No Intelligence.

[Yet Another Anonymous coward]

Re: None of the money makes sense

But suppose one day an AI could autofill TPS reports - it would allow me to fire all those expensive employees with their pensions and health care demands and replace them with a simple SAAS bill

>Remember AI it is just a sycophantic auto complete plagiarising machine. No Intelligence.

You mean it could replace middle managers aswell ?

[Anonymous Coward]

Re: None of the money makes sense

I don't think it is as sycophantic these days as it was 2 years ago...the problem is the opposite now, active refusals to do things that it has been told are off limits based on another AI interpreting your prompt and context history.

[Clausewitz4.1]

Re: None of the money makes sense

AI has good use cases. Problem it isn’t a trillion dollar industry. Way less. Companies will crash and society will adapt. NVidia will do well.

[BoHu]

Good place to be in

It's great to be able to have this debate on the competitive benefits and drawbacks of seeing hefty models of languagerie either keeping their chunky weights close to their chests, or spreading them about openly for all to appreciate, imho.

ISTM though their eventual utility rests largely on how they're kitted out with fashionable agentic girdles ATM, and in that respect the see-through differences between open and close apparels may not be as striking. Or as Matt articulated in the recent Kettle: "harnesses become important and the models actually don't matter" ... [ https://www.theregister.com/ai-and-ml/2026/08/03/the-ai-bubble-is-already-popping-we-just-dont-know-it-yet/5282004 ]

One thing's for sure, it'll be safer, less expensive, and more sustainable for all involved, to execute these procedural agents without a portly model around ... less splatter! ;)

[Ashto5]

Where is the sweet spot

We use these things to do helpful work

Surely we need specific models for specific tasks

Those should be tuned for speed but do we really need 98 quadrillion parameters

We need good testing so that a top 10 for each area can be compiled

Lawyers

1. Parasite 2.0

2. Bludsuck v6.66

You get the idea with simple pricing

[wolfetone]

There is a lot of McCarthyism v2.0 with this. 'murican AI is Freedumb and Chyna is bad m'kay. For all of the talk of China being a dictatorship and America the land of the free, I don't see China blockading ports and blowing up countries because their mate said so. I can't remember the comedian, but he compared China and America as neighbours who would hit your kids, except China are fairly clear cut on that. America pretend they wouldn't.

Anyway. For me I am a big fan of the open weight AI models compared to the closed shop ones and this is based on using them both. I have done so much more with Qwen and DeepSeek compared to Claude/ChatGPT for so much less, it's nonsensical to me to even consider using Claude/ChatGPT. They're expensive and they're not that good compared to what else is out there. And with the squeeze on everything, I have to get the biggest bang for buck. Anthropic/OpenAI are not providing that, and they know that. That's why they're resulting to fearmongering.

As an aside, and I'm surprised no one has pointed this out, remember when the USA blocked China from all of these new technologies etc? Non-Chinese companies coming out with really performant silicone while China lagged behind? Isn't it right to say that this act has helped cause this influx of open weight EFFICIENT LLMs from China because they don't have the infrastructure really to do what the American companies are doing? They're cutting their cloth accordingly, and they are (as we are) reaping the rewards of this political restriction.