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

EuroLLM: LLM made in Europe built to support all 24 official EU languages

eurollm.io

Post reply on HN
Print view View original

301–310 of 629 posts

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#301

Post by notahacker »

runarberg wrote: post #295
purrcat259 wrote: post #16

Earlier quoted context omitted.

I read, write and speak Maltese, AMA if you are curious about the language.

Is there any dialect of Arabic which you can understand without too much effort? How much do you consider Maltese its own language (as opposed to a dialect of Arabic)?

I know that the reverse understanding isn't too bad from chatting with a Saudi-born member of staff on holiday in Malta.

I don't think anyone would seriously consider it a dialect of Arabic though with its completely different alphabet and half the vocabulary and morphology coming from Italian languages/dialects, even if Malta hadn't spent the best part of a millennium trying very hard not to become part of the Arab world

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#302

Post by array_key_first »

whimsicalism wrote: post #242
dzikimarian wrote: post #234

Earlier quoted context omitted.

I took part in application for EU grants a few times and our company group did it many times over the years. It's bureaucracy, often bordering with stupidity. You may need advisors to navigate all their forms & processes. But it certainly isn't "pals-only" type of deal. On the other hand - is it harder than getting VC funding? For seasoned founder with reputation - probably. For fresh startup - probably not.

> For fresh startup - probably not. highly doubt, the whole thing about the success of the US west coast is that they are&were willing to fund unproven upstarts.

Right but if we do this with public funds then the narrative shifts to "OMG the EU is so corrupt and stupid, looking they're pouring taxpayer dollars into unproven stuff! They're deindustrializing!!"

The point being that, as soon as public dollars are on the table, people expect perfection. Anything less is waste, fraud, and abuse.

There's literally no winning. Want to make sure the money is allocated right? Bureaucracy. Want to not do that? Waste, fraud, and abuse.

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#303

Post by OptionOfT »

yvdriess wrote: post #179
ChrisMarshallNY wrote: post #84

Earlier quoted context omitted.

Flemish? I remember watching a TV show in Flemish ( Hotel Beau Séjour [0]), so it's prevalent enough to invest that kind of money into. What about Basque? Is that too controversial? [0] https://en.wikipedia.org/wiki/Hotel_Beau_Séjour

Flemish is more of a political construct than linguistic, it's a grouping of belgian-dutch the coastal, brabant and limburg language groups with each having their own regional dialects.

It's more than political. In speaking Flemish is to Dutch as UK English is to US English. In writing however there is no difference in spelling, but there is a difference in word choice.

Now, being from Belgium, even within that small part of the country where everybody is supposed to speak Dutch, I genuinely don't understand people from near the coast, which was about 150 miles from where I used to live.

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#304

Post by DrNosferatu »

DrNosferatu wrote: post #56

1. It's a nice start, but the EU has to scale to Manhattan Project levels in order to properly compete with the US and China. 2. A credible scale effort for EU own silicon for AI Compute, wouldn't hurt either. 3. And this can only be achieved by vertical integration to combat fragmentation.

Allow me to elaborate,

EuroAI: Europe’s Moonshot to AI Sovereignty

https://open.substack.com/pub/ifiwaspolitical/p/euroai-europ...

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#305

Post by svobodovic »

RobotToaster wrote: post #118
lm28469 wrote: post #24

Earlier quoted context omitted.

Meh, it depends a lot on the dataset, which are heavily skewed towards the main languages. For example they almost always confuse Czech and Slovak and often swap one for the other in middle of chats

Aren't they about as different as American English and British English?

The difference ia larger than let's say just a "dialect". They really are different languages, even though we generally understand each other quite well (younger generations less so). I've heard it's about as different as e. g. Danish and Swedish - not sure if that comparison is helpful.

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#306

Post by nxor »

purrcat259 wrote: post #246
nxor wrote: post #60

Earlier quoted context omitted.

How are loan words viewed? Do businesses work in Maltese? Are monolingual speakers of the language regarded differently than those fluent in English? Do young people in Malta listen to Maltese music?

Maltese has been loaded with loan words since forever. 5 points if you can guess where bonġu, bravu and mappa come from. At some point there was some literary council for the language that decided that any new loan words should just be spelled phonetically. Computer became kompjuter. Businesses do work in Maltese and English. Both are official languages. Its quite rare to encounter a business that deals near exclusiv…

Interesting, but I get the impression that ubiquitous English loan words in seemingly every language is a lot different than loan word patterns of the past. Do you think? Maybe not?

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#307

Post by voxgen »

tensor wrote: post #109
loandbehold wrote: post #10

Aren't all frontier models already able to use all these languages? Support for specific languages doesn't need to be built in, LLMs support all languages because they are trained on multilingual data.

No, that's not how training works. It's not just about having an example in a given language, but also how many examples and the ratio of examples compared to other languages. English hugely eclipses any other language on most US models and that's why performance on other languages is subpar compared to performance on english.

Ratio/quantity is important, but quality is even more so.

In recent LLMs, filtered internet text is at the low end of the quality spectrum. The higher end is curated scientific papers, synthetic and rephrased text, RLHF conversations, reasoning CoTs, etc. English/Chinese/Python/JavaScript dominate here.

The issue is that when there's a difference in training data quality between languages, LLMs likely associate that difference with the languages if not explicitly compensated for.

IMO it would be far more impactful to generate and publish high-quality data for minority languages for current model trainers, than to train new models that are simply enriched with a higher percentage of low-quality internet scrapings for the languages.

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#308

Post by notahacker »

bjourne wrote: post #282
alecco wrote: post #213

Earlier quoted context omitted.

He is 100% right on this one. From personal experience trying to figure out EU. Lawyer bureaucrats manage funds behind red tape clearly meant to be for their pals. All these while the EU is running out of funds and in a process of de-industrialization. There should be an independent corruption investigation on Brussels.

Of course there is red tape. EU funding comes from taxpayer money and we want it to be spent wisely. The red tape is precisely to prevent it from being funneled to pals. EU has funded quite a few free software projects so it's not like the red tape is an insurmountable burden: https://www.ri.se/en/news/blog/europes-digital-future-spells...

I'd also say that their grants aren't unusually burdensome and grantmaking is arms length compared with a lot of other bodies.

Yes, some of the questions are weird, but I'd really rather write a bit confirming that the AI system being developed isn't going to be racist or Skynet than jump through some other hoops that exist (and that absolutely includes VC due diligence). The actual biggest issue with European funds is they get way more competent applications than they can fund anyway.

Re: EuroLLM: LLM made in Europe built to support all 24 official EU languages

#310

Post by Muvasa »

handelaar wrote: post #298
runarberg wrote: post #293

Earlier quoted context omitted.

Is English a legacy official language then from the time the UK was a member (I‘m guessing Ireland nominated Irish instead of English). Aside it feels very un-EU to push this limitation, as I was under the assumption that EU was all about celebrating (European) diversity.

Still an official language, thankfully. Officially, because of Cyprus.

Malta and ireland
Post reply on HN