[embedding-shape]

Sorry if I was unclear, I didn't want to give the impression I think translations or even transcriptions in some cases is easy, or without problems, or not painstakingly time-consuming, it very much is.

I just think building a LLM from scratch is ever harder, with more potential problems that are harder to solve, more time-consuming and even more resource-intensive.

[jononor]

It would require an investment, but those will pay dividends later, as it becomes easier to train LLMs on/for Norwegian. If we need to translate everything to English we might as well just drop using Norwegian altogether. Practically everyone speaks English fluently already...

[embedding-shape]

> as it becomes easier to train LLMs on/for Norwegian

Why would it be easier in the future? The advances we see with LLMs today require a huge amount of data, and it's getting hard getting the amount of data just using any language, I'm having a hard time seeing how it'd get easier for Norwegians to build their own LLM, unless they seriously start to ramp up how much Norwegian content they're putting out.

> If we need to translate everything to English we might as well just drop using Norwegian altogether. Practically everyone speaks English fluently already...

Yeah I mean with that black and white perspective you can pretty much do anything and it won't matter for anything :) I think for the rest of us, what we speak daily and what we rely on professionally, can differ, and that's OK. But maybe this is just my broken Swedish mind being so used to using English professionally but then conversing in Spanish outside of work daily, YMMV.

[jononor]

These models will never compete with frontier models and do not need to - it is about hitting a good-enough, not being the best. Behind the frontier, getting to a certain performance level, is getting easier over time - both sample and compute efficiency is going up.

Furthermore one can reuse investments in data (both agreements, infrastructure and datasets), compute (GPUs, servers) and know-how (training scripts, experienced engineers).

[embedding-shape]

But are you seriously under the belief that all of that, plus all the other things you're forgetting about, is easier, cheaper and faster than transcriptions and translations?

I understand and agree building the LLMs yourself comes with more benefits, long-term ones especially, but still it's harder, more expensive and really time consuming work.

[jononor]

I do not know which is easier. I am not sure that is even well established in research for generative text tasks whether a translation-first or native-language-first is the most sample efficient?

But for a national lab I think it is money well spent to figure out the possibilities and limitations of a native-language LLMs for languages with order of 5M-10M speakers.