Should all countries launch their own LLMs?

While you might know beautiful Chile for its copper, its wines, its pisco-fight with Peru, or its ridiculously long and skinny profile like it’s spooning Argentina (nena, wake up), there’s now a new reason: Chile just launched Latin America’s own large language model. 

Creatively named LatAm-GPT, the idea is to “develop capabilities in the region to have independence and make decisions about how this technology impacts society.”

It took the mateos (slang for nerds) over at the National Center of Artificial Intelligence of Chile two years and two dozen partnerships to develop, but you only need its opening sentence above to see the bigger debate on AI independence. 

The buzzword in AI circles is ‘sovereign AI’, and it’s rapidly evolving from a niche idea to a strategic must, as the world grapples with what’s really at stake with generative AI (beyond market share):

• First, there’s the loss of jobs (though a new study argues AI could be intensifying work, not replacing it)
• Second, there’s the loss of human agency and connection (though again a recent study argues AI helps more than it harms)
• Third, there’s the loss of democratic control (though Taiwan has shown how AI could help de-polarise societies), and
• Fourth, there’s the loss of technological control over super-intelligent systems (though there’s rare US-China agreement on humans controlling nukes, for example).

So with a growing governmental fear of getting stuck in a dependence loop as China and US-based hyperscalers duke it out for dominance, plus a growing user fear of foreign-trained AI erasing folks’ socio-linguistic roots, responses like LatAm-GPT feel inevitable.

And in fact, we’re already seeing them around the world, starting in…

1. The Middle East 

When you have Scrooge M^cDuck piles of cash you can go early — the Emiratis launched Falcon way back in 2023, and their Saudi rivals launched ALLaM the following year.

• The multilingual Emirati model performs well against global benchmarks, whereas
• The Arabic-first Saudi model prioritises regional dialects and cultural nuances.

But they’re both regional sovereignty plays emphasizing Arabic excellence, curbing their foreign dependencies, and all still built on open-source foundations.

Then at the other end of the ‘go early’ spectrum, there’s (ahem)…

2. Europe 

It’s easy to blame the EU’s slower AI approach on its famed AI Act, a kind of ‘sovereignty by standards’ play that emphasises consumer and citizen protections. But while that law might’ve tapped the brakes, Europe’s gentler AI pace predates its regulation for various reasons — think limits around (say) compute, capital concentration, and energy.

Anyway, Europe’s frontrunner is France’s Mistral, founded back in 2023 by ex-DeepMind and Meta researchers. Seen as Europe’s OpenAI, its efficient, open-weight offerings have become popular among European firms wanting to de-risk abroad.

And just yesterday, Mistral announced its first major infrastructure outside France with a new data centre in Sweden, popular for its cheap renewable energy and cooler climes.

Then at the other end of the ‘cooler climes’ spectrum, let’s finish in…

3. Asia

Singapore’s government-backed SEA-LION model is on-brand for the region’s long-time hub — it’s both multilingual (Bahasa, Thai, Vietnamese, Tamil, Khmer, Lao, et al) and multimodal (text + image, with video and audio en route).

By promising more accurate, localised responses, it’s giving neighbours a way to avoid going full-US or full-China, enticing them full-Singapore instead. Again, very on-brand.

But while Japan’s Rakuten is wowing locals with its cultural-linguistic depth, and India’s Sarvam is really flexing its breadth (20,000 dialects!), the region’s top-performer on sheer metrics might be South Korea’s Naver, now enjoying viral adoption, government-backed scale, and legit real world performance.

And with that, maybe go touch some grass while we reflect on what this all means. 

Intrigue’s Take

So many big questions flow from all this, but here are two. First, should every country now develop its own AI? Maybe that depends on how you want to define success, but the presence of so many early, state-backed LLMs suggests many capitals would rather not wait for the answer, or already know the answer: ie, perhaps success here is simply any competent model not dominated by China or the US.

And that leads us to the second question, whose best answer might’ve just been articulated by one of AI’s chief architects, Dario Amodei (CEO of Anthropic): why are so many capitals already charging ahead? Amodei’s latest essay poses a gloomy thought experiment now doing the rounds: imagine a new country appears overnight, populated by 50 million people each way smarter than any Nobel laureate, capable of making ten moves for every one step we take. How should we respond?

Amodei answers his own question like this: “a report from a competent national security official to a head of state would probably contain words like ‘the single most serious national security threat we’ve faced in a century, possibly ever.’”

Now sure, maybe that’s just classic capitalism — there’s a fun parody likening Amodei to a hot-dog CEO warning your only survival option is… buy more hot-dogs.

But enough capitals are seeing something over the horizon or inside the wire, it reminds us of that Oppenheimer line as the world realises the power of nuclear weapons: “I don’t know if we can be trusted with such a weapon. But I know they can’t.”