Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Open Euro LLM: Open LLMs for Transparent AI in Europe

openeurollm.eu

Post reply on HN
Print view View original

261–270 of 289 posts

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#261

Post by closewith »

lxgr wrote: post #256
closewith wrote: post #204

Earlier quoted context omitted.

I stand corrected on this point, the language packs now auto-download. It's still a niche feature only partially built by a Horizon project (it was almost entirely built by commercial entities - MS and Mozilla) in a niche browser. It's an indictment of the Horizon programme that this is considered the pre-eminent success story.

You really can't win with some people. If the EU invests into research and development of a feature that a US tech company already offers (as a proprietary, closed-source service), it's needless duplication and a futile effort in catching up. Yet if it doesn't, that's admitting defeat in the face of competitors and the wrong move as well.

> If the EU invests into research and development of a feature that a US tech company already offers (as a proprietary, closed-source service), it's needless duplication and a futile effort in catching up.

The example we're talking about is powered by a Marian, developed and open-sourced by a US multinational, Microsoft.

The Horizon project was to use that to create a Firefox plugin, which they did.

Another US multinational, Mozilla, later integrated into Firefox.

Firefox has 4.55% market share in Europe.

> Yet if it doesn't, that's admitting defeat in the face of competitors and the wrong move as well.

You are presenting a false and frankly bad faith dichotomy.

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#262

Post by whimsicalism »

jpdus wrote: post #136
FanaHOVA wrote: post #30

Earlier quoted context omitted.

The problem is that: - These are not really super computing cluster in LLM terms. Leonardo is a 250 PFlops cluster. That is really not much at all. - If people in charge of this project actually believe R1 costs $5.5M to build from scratch, it's already over.

I think no one believes that R1 costs $5.5m from scratch. People in this project (most, not all) are very aware of the realities in training and are very well connected in the US as well. Besides Leonardo there are JUWELS, LUMI & other which can be used for ablations and so on. This will never compete with what the frontier labs have (+ are building) but might be just enough for something, that is close enough to be…

what are you all talking about? most people in the industry do believe the publicly stated numbers for dsv3

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#263

Post by closewith »

lxgr wrote: post #255
closewith wrote: post #186

Earlier quoted context omitted.

That would be expected for a system launched 33 years later, but in Galileo and GPS are identical for civilian use (and obviously no-one knows the military capabilities of Block III satellites as that's undisclosed). GPS+Gailleo is the current SOTA, but it's nonsense to say Galileo is "best".

Galileo has signal authentication, GPS doesn't. In a world where GNSS spoofing is increasingly becoming a hazard to aviation and other applications, that's arguably critical.

Not for civilian use, no, although GPS does for military users.

Also, OSNMA is not SA yet.

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#264

Post by tmnvdb »

machiaweliczny wrote: post #259
tmnvdb wrote: post #252

Earlier quoted context omitted.

Developers are cheaper but not a lot, and regulatory costs are higher - the difference in the end in not huge. Plus: if you want to win a global market you need global talent at global wages. You need a lot of private capital and Europe does not have any.

EU has enough capital it just works on US market cause there no unified capital market in EU. If Trump hits EU with tariffs I hope this changes.

You're right, the EU does have a lot of capital, but its not made available to European business.

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#265

Post by Yizahi »

varjag wrote: post #155
ShareTheVista wrote: post #143

Earlier quoted context omitted.

Can you tell us about a project like this that worked?

The Large Hadron Collider clearly works and is infinitely more complex than yet another language model.

The LHC indeed works, as a device. But what was it's purpose and did it do "meaningful" stuff over its lifetime? LHC is a yet another collider which basically can be boiled to the simple idea:

Some scientists invent hypotheses with no basis in reality, and say that they can be proven with a big and expensive collider. When that collider is being built and fail to find the requisite particles, those same scientists say they need a bigger and more expensive collider. GoTo 01.

A lot of these scientists fail to explain what will happen when their hypotheses won't find the requisite particles, essentially generating meaningless papers which are blind stabs at reality. It's like saying that leprechauns exist, but to see them we need a 100 billion euro device. And if we do build it fail to see it, then whoops, it wasn't enough.

tl;dr: LHC is not a particularly good example of proper scientific achievement. More like an achievement in PR and budget grants. Per positive scientific discovery produced there.

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#266

Post by Jon_Lowtek »

Kuinox wrote: post #167
varjag wrote: post #152

Earlier quoted context omitted.

Europe runs its space programme in this way and so far it has pretty good track record. There are more ways to build stuff than the worship of personality.

I don't think Ariane have a "pretty good track record". Most expensive bug of all time that crashed a whole rocket, because of outdated and wrong software engineering practice. They dont innovate, looked down on SpaceX, they have bet against Falcon, and lost the bet. Now they are betting against the Starship. > Honestly, I don’t think Starship will be a game-changer or a real competitor -- ESA chief 2024 https://spac…

> most expensive bug of all time that crashed a whole rocket

being valued at $ 370 million in 1996 that bug was recently dwarved by crowd strikes multi-billion-dollar disaster in 2024

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#267

Post by tessitore »

I see plenty of pessimism in the comments, & talk about how unqualified the organizations people are expected to be, without addressing how important this initiative is to EU & how necessary it is they succeed.

USA has just proven they are economically unpredictable & so unstable they have become fiscally volatile with control in the hands of the lobbiests. This is why Open LLM has the starting support it does already, & for soverign nations is seen as mission critical so as to avoid long term digital services taxations being leveraged like tarrifs against anyone who does not cooperate with whomever is leading the USA.

So to me it feels as though the project is impressive, & quite likely to succeed where others have failed because so few do understand the technology enough to get in the way of progress towards openly standardizing decentralization of AI compute across soverign cloud infrastructures. Even if Aleph Alpha is not able to lead development fast enough, organizations such as OUMI (Open Universal Machine Intelligence) will be working alongside them in attempts to build out the Linuxs of AI frontier modelling.

If nothing else, Open LLM guarantees a raise in the social standards of what it takes to succeed in AI long term. At worst it provides a measure bar of success of global AI innitiatives to be compared against while introducing new organizations & people to the open source ecosystem who would never of otherwise invested in it without the EU stamp of approval.

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#268

Post by wutisgood »

closewith wrote: post #216
rickdeckard wrote: post #208

Earlier quoted context omitted.

> Point at a single project of this sort that had any tangible output that's still in use. Not sure what is your limiting factor (just universities + industry consortiums or explicit IT projects?). Graphene Flagship might be an example, with their research on Graphene they contributed to the foundation of more efficient batteries and solar panels, innovation in automotive and commercial products and so on. Clean Sky…

Graphene Flagship was an irredeemable disaster. €1 B pumped into research and commercialisation and the result is a DOA graphene industry in the EU, left in the dust by China and the US. Clean Sky Joint Undertaking was also a disaster that missed all its targets. As stated elsewhere, CERN is a cautionary tale - the LHC is a vestige of a time when Europe was an economic and scientific powerhouse. If the last 15-20 yea…

[dead]

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#269

Post by varjag »

Yizahi wrote: post #265
varjag wrote: post #155

Earlier quoted context omitted.

The Large Hadron Collider clearly works and is infinitely more complex than yet another language model.

The LHC indeed works, as a device. But what was it's purpose and did it do "meaningful" stuff over its lifetime? LHC is a yet another collider which basically can be boiled to the simple idea: Some scientists invent hypotheses with no basis in reality, and say that they can be proven with a big and expensive collider. When that collider is being built and fail to find the requisite particles, those same scientists sa…

LHC had confirmed the existence of Higgs boson, essentially validating then-hypothetical mechanism for emergence of mass. It also had ruled out a very popular, theoretically beautiful but as it turned out invalid direction of physics studies. You can argue whether it was worth the cost but it was a genuine, significant experimental achievement.

It was also certainly not a PR stunt as it was truly an insanely complex piece of engineering. It is one of humanity's largest technological undertakings ever. Projects of this scale are still beyond the means even of the world's richest oligarchs. But it is a fine example of the things you can achieve in an enormous but well functioning multi-national bureaucracy.

Re: Open Euro LLM: Open LLMs for Transparent AI in Europe

#270

Post by alecmev »

closewith wrote: post #134
input_sh wrote: post #114

Earlier quoted context omitted.

Would you consider for example Firefox' completely-local translations a success? Directly funded by Horizon, made by a consortium of 4 European universities, now a part of Firefox?

7 languages and a feature nobody to a rounding error uses? No, I don't consider that to be a research, economic, or technical success. Honestly, I'm not sure if you posted this is support of Horizon or against? The Horizon budget for 2021-2027 is €95.5 billion or ~€15 billion per annum. If a headline "success" is an unfinished implementation of translation in Firefox of a translation engine (Marian) built by the Micr…

Can you, please, share a source for the claims you make about Marian? Specifically, "developed and open-sourced by a US multinational, Microsoft" and "funded by MS" (from two other comments by you)?

It does look like Microsoft is (was) funding the project, and employs one of the authors as head of research at Microsoft Translator, which is great, but all the "seed" funding and actual research happened in EU. Microsoft hired the author only in 2018 [1], while the earliest EU grant was allocated in 2015 [2], and the main paper they published says "it has mainly been developed at the Adam Mickiewicz University in Poznan and at the University of Edinburgh" [3].

[1] https://www.linkedin.com/in/junczys/

[2] https://github.com/marian-nmt/marian#acknowledgements

[3] https://aclanthology.org/P18-4020.pdf

Post reply on HN