Absolutely not, even if Europe was given all of the compute in the world. The issues are much much worse. Starting at "data privacy", continuing with the "EU AI Act" and just an overall mindset of regulation and German angst. Many structures in the EU actively prevent and fight against innovation. Oftentimes more subtle but the consequences stay the same. If you ever had to use one of the "sovereign" AI provides such as StackIT, OVH and the-like, I feel with you. It is just so bad in terms of product and performance, there is no comparison at all with Hyperscalers, and it shows. Eventually, it's a cultural and structural problem along the way and the future here looks horrible.
We can debate the details and implementation but EU legislature is, at least in spirit, trying to protect human rights. What does it say about progress that the same laws that protect human rights also stifle innovation?
> EU legislature is, at least in spirit, trying to protect human rights. What does it say about progress that the same laws that protect human rights also stifle innovation?
As you said, in spirit. In fact the EU’s AI Act is not really human rights legislation. (It exempts military and national-security uses.) Where it comes close, e.g. in seeking to ban facial recognition or social scoring, it does so clumsily.
So in practice, the EU has passed a series of laws that essentially make AI a monopoly of military and intelligence-community interests while forcing its consumers to use foreign products. Not exactly a win.
> EU legislature is, at least in spirit, trying to protect human rights
Protect human rights as defined by EU legislature, obviously. And privacy in public places, for example, doesn't seem to be an undebatable human right.
Heck, I hate street views disfigured by huge privacy blobs.
> What does it say about progress that the same laws that protect human rights also stifle innovation?
Innovative methods to destroy human life are "stifled" by measures intended to preserve human life. What to you mean by "progress" -- the betterment of the human condition, or the enrichment of the few and powerful?
EU legislature is an actual corpus of laws. It’s imperfect, but it’s arguably better than having a guy that can block a model or threat companies because they crossed him.
The US has a guy who occasionally can screw things up for a few weeks, but who will be gone in a while.
You have it upside down: the innovation and the stuff is the valuable thing, the laws are there to help us organize ourselves a bit after the fact. They're always a secondary concern to the extent that the vast majority of civilization is working with one another, doing material things wherein the law usually is there as a backstop.
There are some ugly things here and there but by and large - 'cookie settings' has not materially improved people's lives - and not nearly as much as the innovations on the web themselves.
Doing is primacy, regulating is always secondary, with only a few exceptions.
The EU is in really really bad shape on industrial issues on a continental scale - 'too many regulations' is actually not a root cause (it's a big drag, but not root), but it's also not for the most part some kind of advantage.
You see the same thing play out with defence and other things.
Having to beg the US for help with Ukraine, for Patriot munitions, Starlink, advanced intel, for 5th Gen gear, mid range ballistic missiles - it's an existentially disempowering posture.
Human rights won't matter in the areas where the Russians have conquerd or destroyed. Again, here EU/Euro governance issues loom large.
'Do the thing' then as you go along, think about some guardrails or whatever, but the 'do the thing' is the hard part that deserves most of the focus.
You can protect human rights without stifling progress. It's not a "pick one of the above" situation.
The EU can and should reform many parts of its sclerotic laws and bureaucracies. Whether it can do so before it becomes a subservient puppet state which serves as a battleground for competing powers remains to be seen.
It is the innovation and wealth that most successfully protect human rights.
It is no accident that we found slavery morally reprehensible enough to ban it right around the time we devised machines that could do a lot of the slave work. Same with the most basic stuff. More lives have been saved by the Haber-Bosch process and dwarf wheat than any legislation.
The idea that cookie banners or attached bottle caps are somehow deeply motivated by the love and protection of human rights is just silly.
> What does it say about progress that the same laws that protect human rights also stifle innovation?
Perhaps it merely says that certain good positive things stifle other good, positive things?
Having 24 languages is a good, positive thing for the EU's cultural distinctiveness, respect for citizens' heritage, and the fairness of the nexus of power not excluding speakers of any country's language.
And yet it's a major barrier to cross-border trade, military cooperation, popular support of closer political ties, and the prospects of any EU companies growing large enough to counterbalance the amazons and facebooks of the world.
A ban on cracking eggs serves the interests of eggs, while stifling the omelette industry.
> We can debate the details and implementation but EU legislature is, at least in spirit, trying to protect human rights
That's an unfounded assertion. Of course, politicians will claim this to be the case. I don't see how patronising citizens protects their human rights, though.
It's certainly a good question. On the idealistic side it's the right choice, people should have the right to have a say in their own data since it's implicitly copyrighted. GDPR has done wonders to prevent careless personal data leaks that are so common in the US, and other kinds of abuse.
In a more practical view though I'm not sure if it'll do anything to stop job replacement from automation as such. Most corporations seem all to eager to make deals with OAI or Anthropic here anyway, and if not that it'll be Chinese ones.
There is a question of "representation", like if a model cannot be trained on the data of one specific country with a specific language, then it does not learn it and the people of that country are now at a disadvantage when trying to leverage the result. Maybe that's a good thing, maybe not, depending on the perspective of how the model is being applied relative to the average person. If it's something that makes their job easier then it's a negative, if it's used by the government to automate scanning all chats then it would be beneficial for it to suck. For widespread languages that doesn't apply of course, so the UK and Spain might as well be exempt.
In general I think it's good for the EU to try and slow down adoption of bleeding edge tech so the US population with its lack of regulations can act as guinea pigs and absorb most of the early damage until we figure out what is the best approach when we get around to adopting it. Even if that means missing out on potential early upsides too. An old example is lots of late adopters going straight to gigabit fiber instead of being stuck on copper DSL.
You can only make money if you produce value. If you stop producing value because your old industries decline/collapse and you miss out on new industries, your wealth will trend towards that of a third world country.
The current question is whether being the minor partner in a relationship with the US or China makes for a better & more free society.
Westphalian norms are falling apart and countries will need to make hard tradeoffs about how to build enough economic heft to maintain their cultural values.
Do you think that companies and States will use "usable" models from Mistral (who has left the race anyway), or frontiers/near frontier ? It's akin to say that walking a good enough means of transportation and we don't need anything else. One day the guys in charriots will remind you of the dire reality.
> And YannLeCun decided to build in Europe as well
AMI has offices across the world. Fact is, companies often have at least an office where the CEO lives. Same when Musk kicked up a stink about 'leaving California'. It wasn't really anything of substance.
You can grumble about the way things are but Europe being so far behind a technological race with important geopolitical ramifications means that you guys are cutting off your nose to spite your face. Have fun with zero of European regulations impacting frontier development and then eventually having to depend on it.
The question was never do enough computers exist in Europe, but rather can Europe organize the capital and cross-company / cross-border relationships required to build a big model at scale. There the answer still looks iffy at best. This is where the US has, and continues to, thrive and where Europe can’t get out of its own way.
I don't believe Europe can build models that can compete with American ones.
1. In order to build competitive models, you need to offer competitive salaries and equity. Europe has very old and inadequate corporate law virtually everywhere but in UK. European labs cannot attract talent that can build such competitive models with the proper lack of incentives.
2. In order to build competitive models, you need gargantuan amounts of compute. And thus capital. How can you compete when big tech can just cough a handful of equity and raise $ 85 B like Alphabet is doing right now?
3. In order to have these datacenters financially feasible you need cheap energy. We don't have it. Some places like France have clean one, but it's still not cheap enough, you're still paying a 45-50% premium over some random South Carolina.
What Europe should do is to finally tackle its fundamental issues with corporate laws, startups and incentivize more money to flow into venture capital.
Essentially we need a bunch of Mistrals, but with more competition and better incentives.
There's plenty of brilliant European engineers and scientists that would gladly take some pay cut to work in Europe instead of US and could bring their expertise here, but you still need the right incentives.
> In order to build competitive models, you need to offer competitive salaries and equity. Europe has very old and inadequate corporate law virtually everywhere but in UK.
It is a fact that a much smaller amount of money is available in the EU for startup investments.
But in which sense it the corporate law inadequate? As far as I am aware the laws allow quite a lot of freedom in setting up the corporate governance for many forms of companies.
You can do what frontier labs do today which is to properly license things that are copyrighted and use open source web crawls for things that don’t have copyright issues. You can then also commission new datasets (volume needed goes down when quality is high).
The European regulations are the thing that will kneecap anything meaningful coming out of Europe. Mind blowing to me that this is worth the tradeoff since Europe will be beholden to other frontier labs be it China or the US, so regulations accomplishing very little if anything on impacting actual AI development and losing vast amounts of leverage in the process.
> You can do what frontier labs do today which is to properly license things that are copyrighted and use open source web crawls for things that don’t have copyright issues. You can then also commission new datasets (volume needed goes down when quality is high).
It cost Anthropic $1.5 billion for training on libgen's 480k pirated ebooks.
Investors will cough up that money if you're already clearly a frontier lab with a model people are paying a lot of money for.
Tough to get that much cash without anything to show.
Regulations aside, Europe is extremely divided. There's constant resistance from individual states, disputes and far right extremism gaining traction. At this point, it seems like EU can barely agree to make any decision.
Well-meaning restrictions that threaten Europe's ability to compete sound like something that would eventually encourage far right extremism by impeaching the validity of the restrictions' philosophical underpinnings.
In my experience it doesn't take anything that ample to encourage far right extremism. It's enough to point at an existing problem and a convincing scapegoat. It works today not eventually and it works regardless of any reason or reality.
It's true though that multiple problems mean multiple propaganda seeds.
> 1. In order to build competitive models, you need to offer competitive salaries and equity.
FWIW, for the equity part there's a proposal expected to pass for next year: https://www.eu-inc.org/ (but it doesn't address taxes, cross-border employment, or anything significant so it's mostly moot). The main goal is to attract native VCs.
AFAIK it's designed by lawyers and old money, with little to no input from tech entrepreneurs.
The culture's still all wrong. If you have a startup that failed in SV, that's par for the course. Better luck next time! If you fail in the EU, that's the kiss of death. Your startup failed???
> If you fail in the EU, that's the kiss of death. Your startup failed???
As a native German, I wouldn't say this is the case: the opinions of people rather differ quite a bit on this topic (there also exist quite a lot of people who are nearer to the US-American mentality here).
What I would rather claim is some kind of kiss of death, is if your startup failed because you made stupid mistakes. People don't like such unresponsible people (many people say that startup founders also have responsibility for their employees).
This sentiment is so wrong, yet so bloody persistent. I know plenty of european entrepreneurs, myself included, that went through one or more screw ups. All the ones that stuck with entrepreneurship eventually landed a hit.
Just as a reminder, Google DeepMind is right here in the UK, and there are hundreds, thousands of top AI researchers that are foundational to everything that the US is doing, as much, if more so than the Chinese contributions.
We need compute, yes, but we certainly aren't short of talent if we put our minds to it, and many of them are already here.
As per your points, Europe really can't compete, particularly when power is considered. However, frontier models that require city-sized data centres might not be all they are cracked up to be.
In China they seem to be nonchalantly doing a lot with AI for specific rather than 'ask me anything' tasks. To them, they are quite used to everyday applications that work well within limited domains, no vast data centre needed, just on-device. Hence the hype is no big deal.
Europe needs to think again about what can be done to make Europe attractive for software development, and I have seen no helpful encouragement from UK or European governments over the last few decades. No word of a lie, all we got in the UK was the BBC Micro, way back in the early 1980s, and since then tech has been culturally uncool.
This cultural aspect has not gone away, if a guy is a software engineer then he isn't going to get lucky with the lasses, they will run a mile.
What gets me is that the UK or places in Europe such as those places where finance matters, could have had active policing and law enforcement of data breaches and hacking, with sensible standards for storing customer data, making Europe the best place to host your data, purely for the legal protections. But we ended up with cookie notices and anti-slavery statements.
It seems unlikely at the moment. The writeup you have is quite interesting, but notably distributed compute for frontier is at least a bump in the road.
As other's have pointed out, if resources across Europe came together there could be a good amount of resources today to do so, but it is fragmented.
Project Glasswing, and most recently the blocking of Fable 5 potentially only for non-US citizens have sparked more attention in the political spheres regarding sovereignty/competitiveness of European AI.
There as also this great write-up which explains the trajectory Europe currently has when it comes to AI compute here: https://europe2031.ai/
So theoretically: yes, but there doesn't seem to be a big enough will to attempt to catch up with the trajectory of American hyper-scalers.
Does Europe need to train a frontier AI model, at all?
"Training a frontier AI model" is a euphemism. The AI industry is, at present, a disproportionately resource-intensive and exploitative activity which carries only an religious promise of making up for itself at some future date (note that the beliefs being touted to board members are different than the gospel spread to users).
More accessible "non-frontier" models are being designed, built, and trained, while bigger models' gains are beginning to asymptote. Economically, Europe does not often hastily participate in new ways of wealth extraction, and with regard to sovereignty, I'd argue that "frontier" models are becoming a liability to the governments who build them and to the populations they're meant to lead.
I'm with you. I think we will need less "frontier" models in the future, but we'll reach for more specialized ones. A small, focused model trained to e.g. coding is within the realm of useful reality. JetBrains is making some moves in the field, and laudably, some of their work is open-weight.
> Does Europe need to train a frontier AI model, at all?
Perhaps. While I'm more impressed by AI than I think you are, I do also say on occasion that recent developments in AI feels much like the 90s rapid development of computer graphics, in the sense we're overly impressed by what we see only to discard it quickly when the next improvement arrives: https://archive.org/details/nextgen-issue-26
If the US government declares no model ≥ Fable can be released, that could make it completely pointless for anyone in the USA to work towards superior models, which may cause rapid catching up, or may cause the investment bubble to pop and all the money to go away.
Or it may cause all the investment to go to not-USA. China's one possibility, EU is another. No idea which would be least-distasteful to the people currently eager to invest in AI.
it is a bizarre nuclear arms race. what they do need are the raw reaources and the capabilities to sefend themselves ans the capability for offense atleast as a threat.
I'm not sure I follow how this metaphor plays out. Many stand to gain by harming and threatening Europe and many stand to suffer if Europe is harmed or threatened. Far be it from me to claim Europe a perfected system, but there are globalized interdependencies which exist here. I think that I'm not seeing the indicators of hard benefits which correlate to the hypotheses put forward by the financial markets. In which ways are the trillions of dollars of value meant to manifest and how do we know when they begin to do so?
No because public EU compute is adminisered by physicists and although they are very competent in their own domains they are winging it when it comes to AI. It like expecting OakRidge to train LLMs...
No, it cannot. The fundamental problem between the US and EU it is that in the US if something is not explicitly forbidden, it is allowed and in the EU if it is not explicitly allowed it is forbidden.
You cannot have innovation at the speed and scale that you have in the US because the legislation is cumbersome and there is no unified market with the same rules.
It's not just that. The US models were trained partly due to massive theft. They illegally acquired (not used, that was legal) massive amounts of texts that they trained their model on. If you or me stole that much we would be in prison but if you're rich you get to do it and pay a small fine. It's not even the laws. It's in the US you're allowed to break the laws.
Sure, theoretically, if it could come to an agreement, and meanwhile the cross-border cooperation of even the inter-EU countries is at an abysmal rate, and currently, even during a goddamn land war with Russia, Germany and France can't come to an agreement to build a fighter jet ensemble (together).
You mean like they've been doing since the 1950s with the the largest physics laboratory in the world (CERN)? Or more to the point, the European High-Performance Computing Joint Undertaking (EuroHPC JU) of which 27/27 EU member states participate?
Even at the smaller level the ethos and ambition is clear - take EURO-3C: the Horizon Europe project aimed at delivering a pan-European sovereign infrastructure that integrates Telco, Edge, Cloud, and AI capabilities under a federated model which has 70+ Euro-participants.
This is unfortunately the problem. The level of the public debate is abysmal, most politicians push unbelivably stupid shit about immigration and other identitarian nonsense, budget gets spent to ensure cheese and wine have the proper AOC certifications on them. Honestly up to a point I even understand it, many people don't see themselves as having a meaningful identity as EU citizens and you can't force it upon them.
Asking for sensible AI policy is like asking for a base on mars.
The fighter jet program was a jobs program, not dissimilar to how many US government programs are jobs programs by having different parts of it made in different states for no goo reason. Add in some nationalism and it was inevitable it would not work out.
I always find it strange that being a "jobs program" is said as if derogatory, especially when talking about military equipment of all things. The free market optimizes for handling normalcy, not exception handling, like wartime. The reason to artificially these jobs is so that you have the factories and expertise to make these things when you need them.
And what, they couldn't split the jobs 50/50 in any meaningful way so that everyone was happy? Dunno, you can let the french design and the germans build or something.
That's how you get something like Ariane 6 -- engineered to satisfy political constraints rather that to be competitive. Granted, NASA or to some extent the US military have the same problem.
In a weird sense, the EU exemplifies what the USA would be like without a strong federal government: Dysfunctional as states compete with, undercut and stifle each other.
isnt the eu basically a less federalist version of the usa? how is it then that the usa claims its more free? the individual state members in the EU have more freedom than individual states.
> isnt the eu basically a less federalist version of the usa?
And what good does it do? The EU cannot speak in a single voice - there is no foreign minister, no defense minister, no whatever minister.
Like the EU foreign function: you have a person claiming to be the EU voice, and then you have the foreign ministers of every EU member that can just say whatever they want if what the EU voice says is contrary to their political game in their own country. Same for the other functions.
Being less federalist is not better, it is worse. The EU does not speak in a single voice in any domain.
The EU is in some ways closer to a confederation, but with various "hacks" - in the form of various treaties - to try to work around the problems that led the US to abandon it in favour of a federal model.
It's largely down to very different levels of willingness to integrate more, and various levels of opposition to the kind of constitutional changes that would be required for a federal model.
Id say the EU is less federal because each member country maintains its own sovereignty and local laws. Its more of a coalition of the willing. While in the US the federal gov can override states. The EU also has each state vote on different legislations. (Massive oversimplification)
The US lays claim to all kinds of bullshit. But more to the point; different values, different laws - does it matter how centralized the control is? I would argue neither is “more” free, just free in different ways. US has a huge problem in equal access to the law, which undermines freedom no matter how good the laws are.
In the US, the freedom is explicitly for the people, not Federal nor State government. States are relatively more free than the Federal government, but both are greatly restricted by the Constitution. Before the Constitution, the US had a previous government under the Articles of Confederation (see below) that had many deficiencies and only gave freedom to member States.
The EU is organized more similar to the US under the Articles of Confederation.
The first American government (Articles of Confederation) gave State governments almost unlimited power e.g. they could print their own money. It did not grant freedom to the people in any meaningful way. States were free to abuse this power both against their own citizens and, more importantly, the other member States. This created many practical problems.
The second and current American government (Constitution) learned lessons from this experience. It removed a limited set of key powers from the States and gave it to the Federal government such as the creation of currency. It also forced all States to interact with each other on the same terms, with strict oversight from the Federal government. Additionally, it explicitly granted rights to the people rather than their State governments, since the States had demonstrated they could not be trusted to do the right thing. These changes forced the States to play nicely with each other and treat their people better.
These changes were a large improvement. Almost every law an American experiences is State law, because States have much more freedom to create laws. The Federal government can only make laws from a short list. Both State and Federal governments are strictly prohibited from creating many kinds of laws.
In the US, freedom is for the people, not for the member States. The Federal government has even less freedom than the member States to make law.
> In the US, the freedom is explicitly for the people, not Federal nor State government.
I don't see how anyone can make this claim in 2026 and not expect to be laughed out of the room? In practice, freedom is for sale; it's not available to the people nor the citizens.
Normal in history. Change often isn't binary. Consider for example: the exact date of the end of the British empire is several possible dates between "the independence of Ireland in the immediate aftermath of WW1" and "it still hasn't".
You don't think US states compete with, undercut and stifle each other?
Second question: you think the USA has a strong federal government?
I mean that is actually an open question even in non-Trump years, not least when one side of the political aisle was famously dedicated to shrinking it down so small it could be "drowned in a bathtub", to quote one of its more famous assholes.
>You don't think US states compete with, undercut and stifle each other?
Compete with each other, yes.
But all US states take pride in Silicon Valley being American.
No EU project for its own "Silicon Valley" can succeed, because the French would insist on it being in France, Germans would insist on Germany, Dutch in the Netherlands, etc.
> No EU project for its own "Silicon Valley" can succeed, because the French would insist on it being in France, Germans would insist on Germany, Dutch in the Netherlands, etc.
France and Germany and the Netherlands may compete with each other to attract VC and tech firms, but they're ultimately no more capable of preventing each other from getting a runaway success loop than New York and Utah were at preventing California from getting actual Silicon Valley.
The EU as an institution just doesn't wield any magic wand of creating things like that to be fought over, just like Washington DC also doesn't.
What is this whole thing about Europe being behind on AI? Do Mistral and DeepL not exist? Yes, I know DeepL is niche, but IMHO it is the best translation model out there.
I've tried using Mistral for various tasks, and it is so far behind the American models that I just never bother using it despite still having lots of Mistral API credits leftover. Even their OCR and TTS products are surpassed by generic US models - I use regular Claude Sonnet for OCR because it is more accurate than Mistral OCR.
I could rant about this, I am just so disappointed at how Mistral completely gave up and pivoted into bespoke fine-tuning consulting. The terrifying thing is that they don't seem to even understand how far behind they are, as if they never tried Opus, let alone Fable / Mythos. Or they do understand and that's why they focus on consulting now.
Well, the front page of their website claims "Frontier AI In Your Hands", so I guess they're not marketing it that way.
I personally think there's a hint in that Mistral Medium 3.5 costs 5x the price of Mistral Large 3, and that Mistral Large is not listed anymore as a "Featured Model" and hasn't been updated since Dec 2025:
But what I really base it on is an interview the Mistral CEO gave on the Big Technology Podcast back in January this year:
Alex Kantrowitz: "Do you consider yourself, is the most important thing you do building the models? Or is the most important thing you do the service? Are you primarily a model builder, or primarily a service provider?"
Arthur Mensch: "We are there to help our customers get to value."
Alex Kantrowitz: "So, service!"
Arthur Mensch: "We are here to... but to get to value, they need to have great models. And to get to value, they need to have the right tools to train the models. And so the best way to train, to create those tools, is effectively to train the best models. So the two things are extremely linked together. We create models that are very easy to customize. We create models with tools that we then export to our customers, so that they can use them, and we help our customers train their own models. You can't go and sell to an enterprise that you are going to help them create great custom systems, if you can't show to the world that you are effectively the leader in open source technology. So the two parts are equally important, the first is enabling the other, and there's effectively a flywheel there because we make our choices when it comes to the model design in a way that is enabling the various customers we have. As one example, we've put a lot of emphasis on having models that are great at physics, because we work with manufacturing companies that run into physical problems. So that's the flywheel we have set up. Having the science team and the business team sit together."
It's at 22:37 in the video. Elsewhere in the podcast he mentions that they don't believe in a large unified generic model, they think the future of AI is small dedicated-task models (OCR, TTS, bespoke trained)... but unfortunately I don't have a timestamp link for that part.
> they think the future of AI is small dedicated-task models
I mean, I do too. Or at least, this will be a large market. But the natural leader for that space is the firm building the frontier models. They have the most control and access for distillation.
It is behind in the sense that if tomorrow the US and China place an export ban on their models all we're left with are Mistral's ones.
They are not bad, and they have made huge progress, but you're still one year behind if not more.
May matter less and less as time progresses, or it may matter more if research further speeds up.
Honestly I wish capitalism and globalization kept working as they did for decades, but since more than a decade we're reverting to inefficient protectionist steps, one after the other.
I feel like the very thing that EU is great at: more consumer friendly and anti-corporate legistation, also hamstrings it for innovation. Why would VCs invest in AI there instead of in the US where they don't have to worry about any of that.
This has factored out product development, which is more than compute resources. Just like any industry, some organisation needs to take ownership and responsibility to convert technology to a usable product.
Just post the prompt that generated this slop next time. Then we'd have a chance of seeing some original thoughts, instead of a bunch of web searches filtered through a bucket of mediocrity.
They could, but why? US and China has poured Trillion of USD into training and any semblance of getting those money back seems like some far fetched dream. Currently there is no realistic path to profitability with these models.
What I see as usable product in the future are smaller specialized models which are able to run and be trained with fraction of resources what goes to current frontier models.
Same problem, different scope. The EU doesn’t have a proper intelligence agencies, it has members who each run their own. The chances that any of them can achieve the scale of China’s spies is silly.
Just bribe someone for access, it's still cheaper than training your own model from scratch. They stole the entire accumulated output of humanity and fed it to their beast, no one should feel any moral or ethical compunction about taking from the AI firms.
Absolutely not, even if Europe was given all of the compute in the world. The issues are much much worse. Starting at "data privacy", continuing with the "EU AI Act" and just an overall mindset of regulation and German angst. Many structures in the EU actively prevent and fight against innovation. Oftentimes more subtle but the consequences stay the same. If you ever had to use one of the "sovereign" AI provides such as StackIT, OVH and the-like, I feel with you. It is just so bad in terms of product and performance, there is no comparison at all with Hyperscalers, and it shows. Eventually, it's a cultural and structural problem along the way and the future here looks horrible.
We can debate the details and implementation but EU legislature is, at least in spirit, trying to protect human rights. What does it say about progress that the same laws that protect human rights also stifle innovation?
> EU legislature is, at least in spirit, trying to protect human rights. What does it say about progress that the same laws that protect human rights also stifle innovation?
As you said, in spirit. In fact the EU’s AI Act is not really human rights legislation. (It exempts military and national-security uses.) Where it comes close, e.g. in seeking to ban facial recognition or social scoring, it does so clumsily.
So in practice, the EU has passed a series of laws that essentially make AI a monopoly of military and intelligence-community interests while forcing its consumers to use foreign products. Not exactly a win.
> EU legislature is, at least in spirit, trying to protect human rights
Protect human rights as defined by EU legislature, obviously. And privacy in public places, for example, doesn't seem to be an undebatable human right.
Heck, I hate street views disfigured by huge privacy blobs.
> What does it say about progress that the same laws that protect human rights also stifle innovation?
Innovative methods to destroy human life are "stifled" by measures intended to preserve human life. What to you mean by "progress" -- the betterment of the human condition, or the enrichment of the few and powerful?
EU legislature is an actual corpus of laws. It’s imperfect, but it’s arguably better than having a guy that can block a model or threat companies because they crossed him.
No it's not.
Europe has no models to even block.
The US has a guy who occasionally can screw things up for a few weeks, but who will be gone in a while.
You have it upside down: the innovation and the stuff is the valuable thing, the laws are there to help us organize ourselves a bit after the fact. They're always a secondary concern to the extent that the vast majority of civilization is working with one another, doing material things wherein the law usually is there as a backstop.
There are some ugly things here and there but by and large - 'cookie settings' has not materially improved people's lives - and not nearly as much as the innovations on the web themselves.
Doing is primacy, regulating is always secondary, with only a few exceptions.
The EU is in really really bad shape on industrial issues on a continental scale - 'too many regulations' is actually not a root cause (it's a big drag, but not root), but it's also not for the most part some kind of advantage.
You see the same thing play out with defence and other things.
Having to beg the US for help with Ukraine, for Patriot munitions, Starlink, advanced intel, for 5th Gen gear, mid range ballistic missiles - it's an existentially disempowering posture.
Human rights won't matter in the areas where the Russians have conquerd or destroyed. Again, here EU/Euro governance issues loom large.
'Do the thing' then as you go along, think about some guardrails or whatever, but the 'do the thing' is the hard part that deserves most of the focus.
You can protect human rights without stifling progress. It's not a "pick one of the above" situation.
The EU can and should reform many parts of its sclerotic laws and bureaucracies. Whether it can do so before it becomes a subservient puppet state which serves as a battleground for competing powers remains to be seen.
The usian 'frontier' "AI" corporations are involved in severe human rights abuses and war crimes.
I'm not so sure the ideal should be to substitute for those.
It is the innovation and wealth that most successfully protect human rights.
It is no accident that we found slavery morally reprehensible enough to ban it right around the time we devised machines that could do a lot of the slave work. Same with the most basic stuff. More lives have been saved by the Haber-Bosch process and dwarf wheat than any legislation.
The idea that cookie banners or attached bottle caps are somehow deeply motivated by the love and protection of human rights is just silly.
> What does it say about progress that the same laws that protect human rights also stifle innovation?
Perhaps it merely says that certain good positive things stifle other good, positive things?
Having 24 languages is a good, positive thing for the EU's cultural distinctiveness, respect for citizens' heritage, and the fairness of the nexus of power not excluding speakers of any country's language.
And yet it's a major barrier to cross-border trade, military cooperation, popular support of closer political ties, and the prospects of any EU companies growing large enough to counterbalance the amazons and facebooks of the world.
A ban on cracking eggs serves the interests of eggs, while stifling the omelette industry.
“Human rights” as defined by the continent that brought you both The Enlightenment and The Holocaust
Feels like a reaaaaal roll of the dice
> We can debate the details and implementation but EU legislature is, at least in spirit, trying to protect human rights
That's an unfounded assertion. Of course, politicians will claim this to be the case. I don't see how patronising citizens protects their human rights, though.
It's certainly a good question. On the idealistic side it's the right choice, people should have the right to have a say in their own data since it's implicitly copyrighted. GDPR has done wonders to prevent careless personal data leaks that are so common in the US, and other kinds of abuse.
In a more practical view though I'm not sure if it'll do anything to stop job replacement from automation as such. Most corporations seem all to eager to make deals with OAI or Anthropic here anyway, and if not that it'll be Chinese ones.
There is a question of "representation", like if a model cannot be trained on the data of one specific country with a specific language, then it does not learn it and the people of that country are now at a disadvantage when trying to leverage the result. Maybe that's a good thing, maybe not, depending on the perspective of how the model is being applied relative to the average person. If it's something that makes their job easier then it's a negative, if it's used by the government to automate scanning all chats then it would be beneficial for it to suck. For widespread languages that doesn't apply of course, so the UK and Spain might as well be exempt.
In general I think it's good for the EU to try and slow down adoption of bleeding edge tech so the US population with its lack of regulations can act as guinea pigs and absorb most of the early damage until we figure out what is the best approach when we get around to adopting it. Even if that means missing out on potential early upsides too. An old example is lots of late adopters going straight to gigabit fiber instead of being stuck on copper DSL.
You don't see how protecting the commons from being exploited by hypercapitalists for their own profit is protecting the rights of the average person?
Some of those seem like good things TBH.
Yes, but they are incompatible with being on the forefront of AI.
Yep. I wonder if being at the forefront of AI makes for a better & more free society.
Being on the backfoot will make us vulnerable to those who are not.
I do not understand how this can not be self evident.
You can only make money if you produce value. If you stop producing value because your old industries decline/collapse and you miss out on new industries, your wealth will trend towards that of a third world country.
Yes there are those two very extreme things and then there is every position in between.
The current question is whether being the minor partner in a relationship with the US or China makes for a better & more free society.
Westphalian norms are falling apart and countries will need to make hard tradeoffs about how to build enough economic heft to maintain their cultural values.
It's not about how large the models are, it's how you use them in society ;)
Not at all, they can just use the petabytes of data sold on the cheap by American and Chinese companies instead of their own.
Bullshit. The AI Act is not the problem, Mistral models are definitely usable and some were/are competitive.
European future looks bright compared to the political landscape the US has now
Do you think that companies and States will use "usable" models from Mistral (who has left the race anyway), or frontiers/near frontier ? It's akin to say that walking a good enough means of transportation and we don't need anything else. One day the guys in charriots will remind you of the dire reality.
I am not denying Mistral is not top tier anymore, (except OCR and STT for speed/efficency) just that the AI Act is not the reason they are not.
But we will see, they have very interesting Enterpise Use Cases, like with Mistral Forge.
And YannLeCun decided to build in Europe as well
> And YannLeCun decided to build in Europe as well
AMI has offices across the world. Fact is, companies often have at least an office where the CEO lives. Same when Musk kicked up a stink about 'leaving California'. It wasn't really anything of substance.
"Innovation"
If your product can't function in a way that respects humans' right to privacy, your product is the problem, not the humans' rights.
You can grumble about the way things are but Europe being so far behind a technological race with important geopolitical ramifications means that you guys are cutting off your nose to spite your face. Have fun with zero of European regulations impacting frontier development and then eventually having to depend on it.
Europe lost religion and gained this instead.
The question was never do enough computers exist in Europe, but rather can Europe organize the capital and cross-company / cross-border relationships required to build a big model at scale. There the answer still looks iffy at best. This is where the US has, and continues to, thrive and where Europe can’t get out of its own way.
I don't believe Europe can build models that can compete with American ones.
1. In order to build competitive models, you need to offer competitive salaries and equity. Europe has very old and inadequate corporate law virtually everywhere but in UK. European labs cannot attract talent that can build such competitive models with the proper lack of incentives.
2. In order to build competitive models, you need gargantuan amounts of compute. And thus capital. How can you compete when big tech can just cough a handful of equity and raise $ 85 B like Alphabet is doing right now?
3. In order to have these datacenters financially feasible you need cheap energy. We don't have it. Some places like France have clean one, but it's still not cheap enough, you're still paying a 45-50% premium over some random South Carolina.
What Europe should do is to finally tackle its fundamental issues with corporate laws, startups and incentivize more money to flow into venture capital.
Essentially we need a bunch of Mistrals, but with more competition and better incentives.
There's plenty of brilliant European engineers and scientists that would gladly take some pay cut to work in Europe instead of US and could bring their expertise here, but you still need the right incentives.
> In order to build competitive models, you need to offer competitive salaries and equity. Europe has very old and inadequate corporate law virtually everywhere but in UK.
It is a fact that a much smaller amount of money is available in the EU for startup investments.
But in which sense it the corporate law inadequate? As far as I am aware the laws allow quite a lot of freedom in setting up the corporate governance for many forms of companies.
You also need to scrape huge amounts of data with no regard for copyright which is:
1. No longer possible the same way it was for openai and anthropic and
2. Much more regulated in the EU
Also the EU would need state backing since we don't have the same private capital, meaning the regulations are even tighter.
You can do what frontier labs do today which is to properly license things that are copyrighted and use open source web crawls for things that don’t have copyright issues. You can then also commission new datasets (volume needed goes down when quality is high).
The European regulations are the thing that will kneecap anything meaningful coming out of Europe. Mind blowing to me that this is worth the tradeoff since Europe will be beholden to other frontier labs be it China or the US, so regulations accomplishing very little if anything on impacting actual AI development and losing vast amounts of leverage in the process.
> You can do what frontier labs do today which is to properly license things that are copyrighted and use open source web crawls for things that don’t have copyright issues. You can then also commission new datasets (volume needed goes down when quality is high).
It cost Anthropic $1.5 billion for training on libgen's 480k pirated ebooks.
Investors will cough up that money if you're already clearly a frontier lab with a model people are paying a lot of money for.
Tough to get that much cash without anything to show.
Regulations aside, Europe is extremely divided. There's constant resistance from individual states, disputes and far right extremism gaining traction. At this point, it seems like EU can barely agree to make any decision.
Well-meaning restrictions that threaten Europe's ability to compete sound like something that would eventually encourage far right extremism by impeaching the validity of the restrictions' philosophical underpinnings.
In my experience it doesn't take anything that ample to encourage far right extremism. It's enough to point at an existing problem and a convincing scapegoat. It works today not eventually and it works regardless of any reason or reality.
It's true though that multiple problems mean multiple propaganda seeds.
> 1. In order to build competitive models, you need to offer competitive salaries and equity.
FWIW, for the equity part there's a proposal expected to pass for next year: https://www.eu-inc.org/ (but it doesn't address taxes, cross-border employment, or anything significant so it's mostly moot). The main goal is to attract native VCs.
AFAIK it's designed by lawyers and old money, with little to no input from tech entrepreneurs.
It had input from entrepreneurs but was disregarded.
The culture's still all wrong. If you have a startup that failed in SV, that's par for the course. Better luck next time! If you fail in the EU, that's the kiss of death. Your startup failed???
> If you fail in the EU, that's the kiss of death. Your startup failed???
As a native German, I wouldn't say this is the case: the opinions of people rather differ quite a bit on this topic (there also exist quite a lot of people who are nearer to the US-American mentality here).
What I would rather claim is some kind of kiss of death, is if your startup failed because you made stupid mistakes. People don't like such unresponsible people (many people say that startup founders also have responsibility for their employees).
This sentiment is so wrong, yet so bloody persistent. I know plenty of european entrepreneurs, myself included, that went through one or more screw ups. All the ones that stuck with entrepreneurship eventually landed a hit.
Just as a reminder, Google DeepMind is right here in the UK, and there are hundreds, thousands of top AI researchers that are foundational to everything that the US is doing, as much, if more so than the Chinese contributions.
We need compute, yes, but we certainly aren't short of talent if we put our minds to it, and many of them are already here.
Deepmind lab is a US company.
By ownership perhaps.
Founded in the UK, then bought by Google, it still has its HQ in London.
As per your points, Europe really can't compete, particularly when power is considered. However, frontier models that require city-sized data centres might not be all they are cracked up to be.
In China they seem to be nonchalantly doing a lot with AI for specific rather than 'ask me anything' tasks. To them, they are quite used to everyday applications that work well within limited domains, no vast data centre needed, just on-device. Hence the hype is no big deal.
Europe needs to think again about what can be done to make Europe attractive for software development, and I have seen no helpful encouragement from UK or European governments over the last few decades. No word of a lie, all we got in the UK was the BBC Micro, way back in the early 1980s, and since then tech has been culturally uncool.
This cultural aspect has not gone away, if a guy is a software engineer then he isn't going to get lucky with the lasses, they will run a mile.
What gets me is that the UK or places in Europe such as those places where finance matters, could have had active policing and law enforcement of data breaches and hacking, with sensible standards for storing customer data, making Europe the best place to host your data, purely for the legal protections. But we ended up with cookie notices and anti-slavery statements.
> all we got in the UK was the BBC Micro, way back in the early 1980s, and since then tech has been culturally uncool.
Off the top of my head, ARM is from Cambridge.
Can Europe train a frontier AI model?
It seems unlikely at the moment. The writeup you have is quite interesting, but notably distributed compute for frontier is at least a bump in the road.
As other's have pointed out, if resources across Europe came together there could be a good amount of resources today to do so, but it is fragmented.
Project Glasswing, and most recently the blocking of Fable 5 potentially only for non-US citizens have sparked more attention in the political spheres regarding sovereignty/competitiveness of European AI.
There as also this great write-up which explains the trajectory Europe currently has when it comes to AI compute here: https://europe2031.ai/
So theoretically: yes, but there doesn't seem to be a big enough will to attempt to catch up with the trajectory of American hyper-scalers.
Thanks! Will give it a read :)
Does Europe need to train a frontier AI model, at all?
"Training a frontier AI model" is a euphemism. The AI industry is, at present, a disproportionately resource-intensive and exploitative activity which carries only an religious promise of making up for itself at some future date (note that the beliefs being touted to board members are different than the gospel spread to users).
More accessible "non-frontier" models are being designed, built, and trained, while bigger models' gains are beginning to asymptote. Economically, Europe does not often hastily participate in new ways of wealth extraction, and with regard to sovereignty, I'd argue that "frontier" models are becoming a liability to the governments who build them and to the populations they're meant to lead.
I'm with you. I think we will need less "frontier" models in the future, but we'll reach for more specialized ones. A small, focused model trained to e.g. coding is within the realm of useful reality. JetBrains is making some moves in the field, and laudably, some of their work is open-weight.
> Does Europe need to train a frontier AI model, at all?
Perhaps. While I'm more impressed by AI than I think you are, I do also say on occasion that recent developments in AI feels much like the 90s rapid development of computer graphics, in the sense we're overly impressed by what we see only to discard it quickly when the next improvement arrives: https://archive.org/details/nextgen-issue-26
If the US government declares no model ≥ Fable can be released, that could make it completely pointless for anyone in the USA to work towards superior models, which may cause rapid catching up, or may cause the investment bubble to pop and all the money to go away.
Or it may cause all the investment to go to not-USA. China's one possibility, EU is another. No idea which would be least-distasteful to the people currently eager to invest in AI.
it is a bizarre nuclear arms race. what they do need are the raw reaources and the capabilities to sefend themselves ans the capability for offense atleast as a threat.
I'm not sure I follow how this metaphor plays out. Many stand to gain by harming and threatening Europe and many stand to suffer if Europe is harmed or threatened. Far be it from me to claim Europe a perfected system, but there are globalized interdependencies which exist here. I think that I'm not seeing the indicators of hard benefits which correlate to the hypotheses put forward by the financial markets. In which ways are the trillions of dollars of value meant to manifest and how do we know when they begin to do so?
No because public EU compute is adminisered by physicists and although they are very competent in their own domains they are winging it when it comes to AI. It like expecting OakRidge to train LLMs...
They already tried training LLMs.
No, it cannot. The fundamental problem between the US and EU it is that in the US if something is not explicitly forbidden, it is allowed and in the EU if it is not explicitly allowed it is forbidden.
You cannot have innovation at the speed and scale that you have in the US because the legislation is cumbersome and there is no unified market with the same rules.
It's not just that. The US models were trained partly due to massive theft. They illegally acquired (not used, that was legal) massive amounts of texts that they trained their model on. If you or me stole that much we would be in prison but if you're rich you get to do it and pay a small fine. It's not even the laws. It's in the US you're allowed to break the laws.
What? Your comment is 100% incorrect. In the EU everything is permitted unless explicitly forbidden.
Source: I am European.
Sure, theoretically, if it could come to an agreement, and meanwhile the cross-border cooperation of even the inter-EU countries is at an abysmal rate, and currently, even during a goddamn land war with Russia, Germany and France can't come to an agreement to build a fighter jet ensemble (together).
You mean like they've been doing since the 1950s with the the largest physics laboratory in the world (CERN)? Or more to the point, the European High-Performance Computing Joint Undertaking (EuroHPC JU) of which 27/27 EU member states participate?
https://en.wikipedia.org/wiki/European_High-Performance_Comp...
Even at the smaller level the ethos and ambition is clear - take EURO-3C: the Horizon Europe project aimed at delivering a pan-European sovereign infrastructure that integrates Telco, Edge, Cloud, and AI capabilities under a federated model which has 70+ Euro-participants.
https://www.medialaws.eu/the-euro-3c-initiative-a-new-dawn-f...
CERN is not an EU project, predates the EU, and is not located in the EU. EuroHPC JU includes a number of non-EU members.
I think that is the way forward: work with whoever has common interests and is willing to work together.
I think the point is that the EU does not necessarily make cooperation between governments any easier.
Compute 2027 coming soon ;)
This is unfortunately the problem. The level of the public debate is abysmal, most politicians push unbelivably stupid shit about immigration and other identitarian nonsense, budget gets spent to ensure cheese and wine have the proper AOC certifications on them. Honestly up to a point I even understand it, many people don't see themselves as having a meaningful identity as EU citizens and you can't force it upon them.
Asking for sensible AI policy is like asking for a base on mars.
The fighter jet program was a jobs program, not dissimilar to how many US government programs are jobs programs by having different parts of it made in different states for no goo reason. Add in some nationalism and it was inevitable it would not work out.
I always find it strange that being a "jobs program" is said as if derogatory, especially when talking about military equipment of all things. The free market optimizes for handling normalcy, not exception handling, like wartime. The reason to artificially these jobs is so that you have the factories and expertise to make these things when you need them.
And what, they couldn't split the jobs 50/50 in any meaningful way so that everyone was happy? Dunno, you can let the french design and the germans build or something.
That's how you get something like Ariane 6 -- engineered to satisfy political constraints rather that to be competitive. Granted, NASA or to some extent the US military have the same problem.
If it's inevitable, then how come the F35's flying around? Sure there's some complaints to be made, but I've seen it work in person.
In a weird sense, the EU exemplifies what the USA would be like without a strong federal government: Dysfunctional as states compete with, undercut and stifle each other.
isnt the eu basically a less federalist version of the usa? how is it then that the usa claims its more free? the individual state members in the EU have more freedom than individual states.
someone explain this to me please
> isnt the eu basically a less federalist version of the usa?
And what good does it do? The EU cannot speak in a single voice - there is no foreign minister, no defense minister, no whatever minister.
Like the EU foreign function: you have a person claiming to be the EU voice, and then you have the foreign ministers of every EU member that can just say whatever they want if what the EU voice says is contrary to their political game in their own country. Same for the other functions.
Being less federalist is not better, it is worse. The EU does not speak in a single voice in any domain.
Sort of, if you squint right.
The EU is in some ways closer to a confederation, but with various "hacks" - in the form of various treaties - to try to work around the problems that led the US to abandon it in favour of a federal model.
It's largely down to very different levels of willingness to integrate more, and various levels of opposition to the kind of constitutional changes that would be required for a federal model.
Id say the EU is less federal because each member country maintains its own sovereignty and local laws. Its more of a coalition of the willing. While in the US the federal gov can override states. The EU also has each state vote on different legislations. (Massive oversimplification)
Because the laws are different? Are you really confused?
The US lays claim to all kinds of bullshit. But more to the point; different values, different laws - does it matter how centralized the control is? I would argue neither is “more” free, just free in different ways. US has a huge problem in equal access to the law, which undermines freedom no matter how good the laws are.
In the US, the freedom is explicitly for the people, not Federal nor State government. States are relatively more free than the Federal government, but both are greatly restricted by the Constitution. Before the Constitution, the US had a previous government under the Articles of Confederation (see below) that had many deficiencies and only gave freedom to member States.
The EU is organized more similar to the US under the Articles of Confederation.
The first American government (Articles of Confederation) gave State governments almost unlimited power e.g. they could print their own money. It did not grant freedom to the people in any meaningful way. States were free to abuse this power both against their own citizens and, more importantly, the other member States. This created many practical problems.
The second and current American government (Constitution) learned lessons from this experience. It removed a limited set of key powers from the States and gave it to the Federal government such as the creation of currency. It also forced all States to interact with each other on the same terms, with strict oversight from the Federal government. Additionally, it explicitly granted rights to the people rather than their State governments, since the States had demonstrated they could not be trusted to do the right thing. These changes forced the States to play nicely with each other and treat their people better.
These changes were a large improvement. Almost every law an American experiences is State law, because States have much more freedom to create laws. The Federal government can only make laws from a short list. Both State and Federal governments are strictly prohibited from creating many kinds of laws.
In the US, freedom is for the people, not for the member States. The Federal government has even less freedom than the member States to make law.
> In the US, the freedom is explicitly for the people, not Federal nor State government.
I don't see how anyone can make this claim in 2026 and not expect to be laughed out of the room? In practice, freedom is for sale; it's not available to the people nor the citizens.
This is irrelevant to the subject at hand. Not everything needs to be about whatever outrage bait is the current thing on your social media feed.
Normal in history. Change often isn't binary. Consider for example: the exact date of the end of the British empire is several possible dates between "the independence of Ireland in the immediate aftermath of WW1" and "it still hasn't".
You don't think US states compete with, undercut and stifle each other?
Second question: you think the USA has a strong federal government?
I mean that is actually an open question even in non-Trump years, not least when one side of the political aisle was famously dedicated to shrinking it down so small it could be "drowned in a bathtub", to quote one of its more famous assholes.
>You don't think US states compete with, undercut and stifle each other?
Compete with each other, yes.
But all US states take pride in Silicon Valley being American.
No EU project for its own "Silicon Valley" can succeed, because the French would insist on it being in France, Germans would insist on Germany, Dutch in the Netherlands, etc.
> No EU project for its own "Silicon Valley" can succeed, because the French would insist on it being in France, Germans would insist on Germany, Dutch in the Netherlands, etc.
France and Germany and the Netherlands may compete with each other to attract VC and tech firms, but they're ultimately no more capable of preventing each other from getting a runaway success loop than New York and Utah were at preventing California from getting actual Silicon Valley.
The EU as an institution just doesn't wield any magic wand of creating things like that to be fought over, just like Washington DC also doesn't.
Why even bother creating a repo like this? Why not just link to a ChatGPT conversation?
And at least save us the pain of trying to scroll through the bs intro. Just put the 10 or whatever bullet points right there.
What is this whole thing about Europe being behind on AI? Do Mistral and DeepL not exist? Yes, I know DeepL is niche, but IMHO it is the best translation model out there.
I've tried using Mistral for various tasks, and it is so far behind the American models that I just never bother using it despite still having lots of Mistral API credits leftover. Even their OCR and TTS products are surpassed by generic US models - I use regular Claude Sonnet for OCR because it is more accurate than Mistral OCR.
I could rant about this, I am just so disappointed at how Mistral completely gave up and pivoted into bespoke fine-tuning consulting. The terrifying thing is that they don't seem to even understand how far behind they are, as if they never tried Opus, let alone Fable / Mythos. Or they do understand and that's why they focus on consulting now.
>Or they do understand and that's why they focus on consulting now.
Cohere would like a word here
Did Mistral really throw in the towel on building frontier models?
Well, the front page of their website claims "Frontier AI In Your Hands", so I guess they're not marketing it that way.
I personally think there's a hint in that Mistral Medium 3.5 costs 5x the price of Mistral Large 3, and that Mistral Large is not listed anymore as a "Featured Model" and hasn't been updated since Dec 2025:
https://docs.mistral.ai/models/overview
But what I really base it on is an interview the Mistral CEO gave on the Big Technology Podcast back in January this year:
Alex Kantrowitz: "Do you consider yourself, is the most important thing you do building the models? Or is the most important thing you do the service? Are you primarily a model builder, or primarily a service provider?"
Arthur Mensch: "We are there to help our customers get to value."
Alex Kantrowitz: "So, service!"
Arthur Mensch: "We are here to... but to get to value, they need to have great models. And to get to value, they need to have the right tools to train the models. And so the best way to train, to create those tools, is effectively to train the best models. So the two things are extremely linked together. We create models that are very easy to customize. We create models with tools that we then export to our customers, so that they can use them, and we help our customers train their own models. You can't go and sell to an enterprise that you are going to help them create great custom systems, if you can't show to the world that you are effectively the leader in open source technology. So the two parts are equally important, the first is enabling the other, and there's effectively a flywheel there because we make our choices when it comes to the model design in a way that is enabling the various customers we have. As one example, we've put a lot of emphasis on having models that are great at physics, because we work with manufacturing companies that run into physical problems. So that's the flywheel we have set up. Having the science team and the business team sit together."
It's at 22:37 in the video. Elsewhere in the podcast he mentions that they don't believe in a large unified generic model, they think the future of AI is small dedicated-task models (OCR, TTS, bespoke trained)... but unfortunately I don't have a timestamp link for that part.
https://www.youtube.com/watch?v=xxUTdyEDpbU&t=1357s
> they think the future of AI is small dedicated-task models
I mean, I do too. Or at least, this will be a large market. But the natural leader for that space is the firm building the frontier models. They have the most control and access for distillation.
It is behind in the sense that if tomorrow the US and China place an export ban on their models all we're left with are Mistral's ones.
They are not bad, and they have made huge progress, but you're still one year behind if not more.
May matter less and less as time progresses, or it may matter more if research further speeds up.
Honestly I wish capitalism and globalization kept working as they did for decades, but since more than a decade we're reverting to inefficient protectionist steps, one after the other.
Under normal circumstances, I'd agree with most folks here, it'd be highly unlikely.
However, we're (i think officially) in an arms race.
I wouldn't want to bet against anyone in these unprecedented times (with plenty of historical parallels).
thanks chatgpt
Clear giveaway: "Honest caveats"
Hey, the point is clarity, not novelty!
Its a README not poetry
Or using GitHub for a couple files
"Three layers."
I feel like the very thing that EU is great at: more consumer friendly and anti-corporate legistation, also hamstrings it for innovation. Why would VCs invest in AI there instead of in the US where they don't have to worry about any of that.
This has factored out product development, which is more than compute resources. Just like any industry, some organisation needs to take ownership and responsibility to convert technology to a usable product.
Just post the prompt that generated this slop next time. Then we'd have a chance of seeing some original thoughts, instead of a bunch of web searches filtered through a bucket of mediocrity.
Just use AI to summarize to the prompt.
You're cute when you're cranky
> Can Europe train a frontier AI model on the compute it owns?
No
They could, but why? US and China has poured Trillion of USD into training and any semblance of getting those money back seems like some far fetched dream. Currently there is no realistic path to profitability with these models.
What I see as usable product in the future are smaller specialized models which are able to run and be trained with fraction of resources what goes to current frontier models.
Who cares? Just distill some existing frontier models and run the inference yourselves. Instant sovereign AI.
Until they choose to lock their models down (e.g. Fable). Building policy off copying isn't reliable.
By then you've already distilled it. Copying is super reliable. It's a historically-proven way to get nearly the same thing at a fraction of the cost.
https://en.wikipedia.org/wiki/First-mover_advantage#Second-m...
Howre you going distill if they block your access e.g. Fable
How you make ROI when your model is blocked?
Chinese are probably laughing reading this.
Same problem, different scope. The EU doesn’t have a proper intelligence agencies, it has members who each run their own. The chances that any of them can achieve the scale of China’s spies is silly.
Just bribe someone for access, it's still cheaper than training your own model from scratch. They stole the entire accumulated output of humanity and fed it to their beast, no one should feel any moral or ethical compunction about taking from the AI firms.
Welcome to the age of digital realpolitik.
They really shouldn’t.
those that know, do not talk?