Europe Must Invest in Open Source AI or Cede To China, Schmidt Says (bloomberg.com) 65
An anonymous reader shares a report: Europe must invest in its own open source artificial intelligence labs and address soaring energy prices, or it will quickly find itself dependent on Chinese models, former Google chief executive and tech investor Eric Schmidt said.
"In the US, the companies are largely moving to closed source, which means they'll be purchased and licensed and so forth. And it is also the case that China is largely open weight, open source in its approach," Schmidt said at the World Economic Forum in Davos, Switzerland, on Tuesday. "Unless Europe is willing to spend lots of money for European models, Europe will end up using the Chinese models. It's probably not a good outcome for Europe."
"In the US, the companies are largely moving to closed source, which means they'll be purchased and licensed and so forth. And it is also the case that China is largely open weight, open source in its approach," Schmidt said at the World Economic Forum in Davos, Switzerland, on Tuesday. "Unless Europe is willing to spend lots of money for European models, Europe will end up using the Chinese models. It's probably not a good outcome for Europe."
Just wait for the bubble to pop (Score:2)
...and then laugh at both 'Murica and Jiiihna.
Re: (Score:2)
Re: (Score:2)
Is there a readable version of the article. Its stuck behind a paywall I have no intention of paying, which is bit of a "I have no idea how to have an informed opinion on this?" scenario.
Re: (Score:2)
On December 10, 2025, the Federal Open Market Committee (FOMC) directed the Open Market Trading Desk (the Desk) at the Federal Reserve Bank of New York to increase System Open Market Account (SOMA) securities holdings to maintain an ample level of reserves through purchases in the secondary market of Treasury bills (or, if needed, of Treasury securities with remaining maturities of 3 years or less). These reserve management purchases (RMPs) will be sized to accommodate projected trend growth in the demand for Federal Reserve liabilities as well as seasonal fluctuations, such as those driven by tax payment dates.
Monthly amounts of RMPs will be announced on or around the ninth business day of each month alongside a tentative schedule of purchase operations for the subsequent approximately thirty days. The Desk plans to release the first schedule on December 11, 2025, with a total amount of RMPs of approximately $40 billion in Treasury bills; purchases will start on December 12, 2025. The Desk anticipates that the pace of RMPs will remain elevated for a few months to offset expected large increases in non-reserve liabilities in April. After that, the pace of total purchases will likely be significantly reduced in line with expected seasonal patterns in Federal Reserve liabilities. Purchase amounts will be adjusted as appropriate based on the outlook for reserve supply and market conditions.
The Desk was also directed in October to reinvest all principal payments from the Federal Reserve's holdings of agency securities into Treasury bills via secondary market purchases. The monthly schedule of planned purchases will include RMPs as well as these purchases.
The Desk plans to distribute the monthly secondary market purchases across two Treasury bill sectors. Purchase amounts in each sector will be determined by sector weights. These sector weights will be based on the 12-month average of the par amount of Treasury bills outstanding in each sector relative to the total amount outstanding across the two sectors as initially measured at the end of September 2025.
So in essence not at all what the poster claimed it was.
oh no (Score:4, Insightful)
So to be clear, it's not good for Europe to depend on open source models they don't have to pay for where someone else is doing the work and spending the energy to create them? Sounds terrible all right.
Re: (Score:1, Flamebait)
Relying on a country you may be at war with before long... what could possibly go wrong?
Re: oh no (Score:5, Insightful)
Anyone who is making themselves reliant on general purpose LLM technology is fucking themselves over no matter who controls it.
Re:oh no (Score:5, Insightful)
How would using open source make one country reliant on another, much less one they "may be at war with"? That's the thing with open source, once you have it you have it.
Re:oh no (Score:5, Insightful)
The Chinese "open source" LLM is a huge binary blob. Who knows what's really hidden inside it?
Re: (Score:2)
Which one specifically?
Re:oh no (Score:5, Interesting)
All of them really. What's typically open source is
1) the code used for training, but never the dataset for initial LLM and never the RLHF (reinforcement learning with human feedback) data used to make a text vomiting LLM into a useful question answering maching.
2) the resulting weights - these are totally uninterpretable.
So it's never fully replicable; even if you had the infra and were willing to burn electricity you don't have a way of going to 2) yourself.
AFAIK that's not just the Chinese but also open-source / weights Llama and Mistral.
Re: (Score:2)
All that you say is true. If you don't want to have inferior LLMs then depending on open sourced LLMs is not a working strategy.
But, and this is a big ol' ass, so what? LLMs do a bunch of cool tricks, but one one hand none of them can be used unsupervised, and on the other even mediocre ones are pretty good. Unless someone can demonstrate that you can hide spyware in weights, you can reasonably evaluate the fitness of foreign LLMs the same way you would a home-grown LLM, i.e. by questioning it on topics tha
Re: (Score:2)
Open-weights is not open-source.
You can rebuild the blob from the source with open source. You cannot do that with any of the open-weight models right now.
The biggest, and most critical thing for me that is missing, is the fine-tuning (supervised and unsupervised) data.
This means your model can be fine-tuned to offer very subtle propaganda with each response that you may never notice individually, and you will never be able to find it, or produce a version without
Re: oh no (Score:2)
Your average schmoe who is unqualified to tell the difference isn't going to ask a question where it matters, and anyone who is in a position where what they get from it matters either shouldn't be allowed to do so in the first place because it might just hallucinate some worse than worthless bullshit, but if they are qualified to detect that then they are also qualified to detect bias. It's a non problem because any interference with the results which would actually matter is easy to detect.
Re: (Score:2)
Re: (Score:2)
Oh yeah, I must be dumb, I forgot you're a classic dipshit for long enough to write you a serious reply. Back to only abuse for you from here on out, I guess.
Re: (Score:2)
Oh yeah, I must be dumb
That is a fair statement. The only person dumber than you on slashdot is rsilvergun, which is quite a low bar.
Re: (Score:2)
A war between Europe and China is geographically impractical.
Re: (Score:2)
A war between Europe and China is geographically impractical.
Not really, the war will be fought on someone else's territory.
Re: (Score:1)
Not really, the war will be fought on someone else's territory.
If Europe is smart, it will be fought in Russia.
If Europe is stupid, it will be fought somewhere between Russia and the heart of the EU. Like, say, Ukraine.
If the US were smart it would probably already be going on in Russia. But alas, we were stupid enough to put a Russian asset in the big chair.
Re: (Score:2)
If Europe is smart, it will be fought in Russia.
Europe's military capacity is nil, it has been since WWII. They've been just relying on the US to do that for them, despite multiple US administrations over the decades encouraging them to build up their own militaries. Trump, due solely to his own sheer incompetence, is the first president to actually convince them to do so, but they have a very long way to go. Any capacity that they can offer that can do anything meaningful to Russia, either offensively or defensively, is 100% provided by the US. Most of
Re: oh no (Score:2)
You're thinking about the old kind of warfare with big machines that get stuck in the mud during advances, you know like Russia at the start of the invasion with their Chinese tires.
Re: (Score:2)
And how else were you proposing that Europe does this? Fly in with a Rafale? Collectively, Europe has some US made Reaper drones, but nowhere near enough to do anything meaningful, and certainly not without deploying them from much closer, or even within Russia as Ukraine currently does. Maybe the EU could demoralize Moscow by bombing the Vodka factories at best, because sure as shit they aren't going to go anywhere nearly far enough into Siberia to take out any meaningful fossil fuel extraction or processi
Re: (Score:2)
Where? There is half a world between Europe and China. China would have to conquer most of Asia before they can come close enough to Europe for a war even in a different country can be practical. Look at the bloody map if you are geographically challenged.
Re: (Score:1)
Re: (Score:2)
Don't be ridiculous. The only reason why it happened was an extreme disparity regarding weaponry. That window has closed decades ago. You can look at the battle of Tsushima to see what happens when a country tries to fight a near peer across half the world.
Re: (Score:1)
The Battle of Tsushima is what happens when typical Russian military incompetence engages with a superior enemy, whether 5 feet across their southern border, or the Sea of Japan.
You're an idiot.
Re: (Score:2)
Yes, the very big disparity of weaponry between the Allies and Germans in Europe and France, Japan and Indochina, the US and the Pacific.
An interesting thing about the Pacific war is that the US had no meaningful capacity to project sea power all the way across the Pacific at the time of Pearl Harbor, but Japan did and already had been. Japan knew this. But Japan was also in denial about the wartime production capacity of the US, which to this day remains the highest the world has ever seen, and has never been matched since. Their own research indicated as such, but they dismissed their sources as propaganda and lies. Fairly quickly into it,
Re: (Score:2)
Where? There is half a world between Europe and China. China would have to conquer most of Asia before they can come close enough to Europe for a war even in a different country can be practical. Look at the bloody map if you are geographically challenged.
Given that you are politically challenged, look at the Belt and Roads initiative and China's Neo-Colonialism in Africa and Latin America, among other places.
Re: (Score:3)
It's problematic for assurance because users cannot tell what undisclosed biases are baked into the weights -- and relying on someone else to do the bulk of the training means you cannot produce your own weights.
Even in classic open source, think of cases like an implementation of Dual_EC_DRBG, a malicious version of gotofail, or the xzutils backdoor. Those are all cases where the source code was available.
Re: oh no (Score:2)
So you mean you would need to test for accuracy like is already done with all models? Oh no! Anyway...
Re: oh no (Score:2)
Your argument depends on thinking weights and code are the same. Please die and make slashdot smarter.
Re: (Score:2)
Re: (Score:2)
Checking for bias is another entirely.
Surely your brain is functional enough to understand that a fact can be given with different points of view.
Re: oh no (Score:2)
If you had a brain you would know that you can absolutely test for bias. For example, you're biased towards being a little bitch.
Re: (Score:2)
Yes, there are ways to evaluate for bias. However, bias is not accuracy. There is no boolean evaluator- there is no ground truth.
Looking for bias is like scanning for viruses. It's an ongoing game of cat and mouse that never ends. It's politics.
Re: (Score:2)
But the developers aren't white people!! The evil Yellow Devil can't be trusted! They're simultaneously sub-human and also somehow able to insert spyware in open source software that can't be detected! /s
(Yes, the upper class of Europe is that racist and that stupid.)
How about no (Score:5, Insightful)
There is no reason to waste money on bullshit, and the world's biggest advertisement peddler is no visionary of any kind.
Re: (Score:3)
Re: (Score:2)
But but but...
Since OpenAI has bought or pre-ordered all RAM and GPUs and CPUs that will be manufactured in the next 5 years, how can Europe do anything??!?
Perhaps it is best to see how China does and buy what's necessary on AliExpress, with free shipping.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
There is no reason to waste money on bullshit, and the world's biggest advertisement peddler is no visionary of any kind.
Yep, a future /. headline from early next year "Europe wisely bankrupting itself on failed AI bubble".
or not use models at all (Score:2)
"Europe will end up using the Chinese models. It's probably not a good outcome for Europe."
This assumes that using these models is mandatory. We don't know that, and it could very well be a "good outcome" for Europe.
Re: (Score:2)
I think the outcome that Tech wants for Europe is to have its economy wrecked simultaneously with America's when the bubble pops.
They don't want people to see that there's a viable alternative to their brand of self-enriching, economy-destroying business.
It's a similar to when the Republicans tear up any kind of government program that is doing people good, simply because a Democrat somewhere might get credit for improving something. It's very much a hostage-taking, terroristic tactic.
naah (Score:3, Insightful)
Schmidt wants Europe to prepare space for Google since the US is becoming less and less stable.
For European companies it makes much more sense to wait, or at least wait until models become better, and then steal whatever is available and start from there. I think it will be a hilarious trial if some European company will defend itself against Google, where Google has to assert that a A) it's Google's copyrighted property and B) Google has not used any copyrighted material from anyone else (which presumably would invalidate the copyright).
Energy pricing (Score:5, Informative)
Re: (Score:2)
Really there should be a requirement that new data centers include new clean power generation to 100% support their power needs. Water for data centers would also need considered
Re: (Score:2)
They could have much cheaper energy if they hadn't allowed the US to blow up Nordstream.
death of open source (Score:2)
Open source business model fails when you give stuff away for free to people who want to bury you.
No money, no energy, no skilled engineers (Score:2)
Europe has been digging its own hole to fall into with its EU overreach.
Or European closed-source... (Score:2)