Live data from Hacker News

Can Europe train a frontier AI model on the compute it owns?

github.com

241–250 of 313 posts

Re: Can Europe train a frontier AI model on the compute it owns?

#241

Earlier quoted context omitted.

I mean US is doing the same thing with intel so not like this is EU exclusive

You cannot be seriously comparing Mistral to Intel. Intel might be falling behind right now, but they were pioneers in 90s and 2000s. When has Mistral ever been a pioneer on anything?

Actually at MoE. Though Deepseek took that approach and improved it by a lot.

Re: Can Europe train a frontier AI model on the compute it owns?

#242
post #22

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 gargantua…

FLUX by Black Forrest Labs would disagree there and it is built right in Germany.

Re: Can Europe train a frontier AI model on the compute it owns?

#243
post #79

Earlier quoted context omitted.

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 a…

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.

It is partly located in the EU and primarily funded by European countries

Re: Can Europe train a frontier AI model on the compute it owns?

#244
post #218

Being slow is sometimes a problem. But if the needed total revenue for an investment is so absurdly large, that we need to capture 5-10% of the labour market, being slow might mean not burning in the crash. The US drove the CCCP to bankruptcy by investing in nuclear bombs. Maybe history repeats itself.

The US was able to do that because a majority of the world including Europe was determined to exclude USSR from all trade and by putting them into extreme paranoia of total destruction. EU has lots of trading partners from the West and not everything is improved by AI. Yes an economy that's being inflated by AI can affect acquiring of certain resources like DRAM or certain silicon quotas, if you're not joining the ba…

I was thinking about china driving the US to bankruptcy by commoditising LLMs, thus reducing the revenue.

Re: Can Europe train a frontier AI model on the compute it owns?

#245
We do. But only for a limited context space, plant images. We do train our own Foundation Model, and it works great. It's the world frontier model. The machine is expensive, yes, but the customers also pay a lot to get it's advantages. And we got the H100`s on ebay.

We also have buerocracy and regulations problems, but not with the AI at all. Only with the military civil dual-use restriction of thermal cameras. The ones we are allowed to sell have ridicolous resolutions. It still works, but it's barely useful.

And our colleagues across town are also world leader in their AI field. They produce all the world best face recognition SW. They had to use tricks to overcome regulations, but it works. They have their own HW as us.

And again other colleagues across town do rent their GPU computes to train their models. That's for office OCR, and speech to text.

The town is Dresden btw

Re: Can Europe train a frontier AI model on the compute it owns?

#246
post #245

We do. But only for a limited context space, plant images. We do train our own Foundation Model, and it works great. It's the world frontier model. The machine is expensive, yes, but the customers also pay a lot to get it's advantages. And we got the H100`s on ebay. We also have buerocracy and regulations problems, but not with the AI at all. Only with the military civil dual-use restriction of thermal cameras. The o…

Where can I get more info on the model, would love to read up about it

Re: Can Europe train a frontier AI model on the compute it owns?

#247
post #224

Earlier quoted context omitted.

It absolutely has. It's always more regulation, more bureaucracy and more red tapes. They are just creating more barriers to keep themselves employed and to expand the reach of their authority.

Actually right now the reverse is happening. GDPR will become easier there is simplification coming, there will be less national goldplating

They have been discussing that for years, but still nothing happened. At the rate of the bureaucracy it would take few more years to even draft the laws to ease GDPR

Re: Can Europe train a frontier AI model on the compute it owns?

#248
post #110

Earlier quoted context omitted.

By ownership perhaps. Founded in the UK, then bought by Google, it still has its HQ in London.

Does it matter? If they develop the "next big thing" and the next US administration decides its for US citizens only, nothing has changed.

Does it matter in which countries the offices with the staff who develop some of these models are? I think so, for a number of reasons too long to go into.

Re "US citizens only", the question I'd have right now for all the AI companies is: can any new better-or-as-good-as Fable AI model actually be released?

If not, stop development of better models. That's the single most expensive thing any of them are doing.

Export controls on software is somewhere between "hard" and "impossible", so "US citizens only" means something between "nobody" and "perhaps some government agents", but the latter don't pay enough to justify the cost.

Re: Can Europe train a frontier AI model on the compute it owns?

#249

Earlier quoted context omitted.

Political landscape of the US right now will change in few years like it has been changing for the last 250 years. USA does not equate to its current political situation. But EU would keep on rotting from inside and I would eat my hat if Mistral doesn’t go out of business in next 5 years

> I would eat my hat if Mistral doesn’t go out of business in next 5 years Hope you're hungry. The Mistral are going to what most great European companies are good at - regulatory arbitrage. They're going to insert themselves everywhere within EU (French govt, etc) and extract value that way whilst delivering subpar services to what open weight Chinese models can deliver. Honestly they'll probably be profitable befor…

It could be true that French government might use it, but I hardly think Germans or Italians or any other government would trust Mistral or use it. Even though they are part of EU, they still don’t trust each other completely when it comes to national security and sharing intelligence. I can easily imagine BND fine tuning some Chinese model or still relying on American models

Re: Can Europe train a frontier AI model on the compute it owns?

#250
post #150

Earlier quoted context omitted.

The founding fathers created a document that was already struggling with modern realities prior to Trump. 250 years is not a particularly impressive amount of time for a country to not fall apart.

Care to elaborate on concrete examples on where it struggled? 250 years is quite impressive even if you don't believe it or not because only a handful of countries in the whole world has an older constitution.

There was a civil war, that killed millions, it was fought over slavery.

250 years is commendable, but it wasn't without problems.

'stable' as we understand it is relatively modern.

Post reply on HN