We live in such unexpected, shifting, fracturing geopolitical times just now. A stability taken for granted for decades no longer seems a given. So much so, that many have begun to question the global tech ecosystem we are embedded in, considering the safety of our data and workflows, and seeking less exposed, closer-to-home alternatives.

It’s something we can explore without straying into conspiracy territory, and it goes beyond data security. Tech writer Cory Doctorow has written at length on the downsides to walled garden platforms that make leaving costs high while degrading (or enshittifying – Macquarie Dictionary’s 2024 word of the year) their services. Linguaphiles should know – our own beloved Duo is one of them. It’s a compelling argument, and one that national consumer protection agencies are starting to incorporate into policy. The notion that we can take meaningful steps to decouple from tech monopolies is beginning to take hold.

Ecosystem creep : AI

This leads us to AI firms – arguably the fastest growing of tech behemoths, whose services nonetheless are working their way into many of our workflows. It’s not all doom and gloom here, though; Anthropic in particular has emerged as one US company willing to stand up for an ethical stance in the field.

That said, most European LLM traffic still goes down that American route, collecting on servers users’ states have no jurisdiction over. Users come to rely more and more on these services for key elements of their day-to-day, although have little control over their place in that ecosystem.

So what to do? LLMs are incredibly useful tools for a number of creative applications. For language teachers, they are particularly good at creating authentic-sounding materials for worksheets. In fact, I’ve often argued that LLMs are a tech almost tailor-made for language learning and teaching – in few other fields is the language structure more important than the actual content! They’re genuinely brilliant at creating copy, often highly nuanced, for learning.

AI Swaps

Well, one quick and easy swap is Le Chat by French AI company Mistral. It’s a ‘full fat’ LLM on a par with the big US names, running your prompts remotely on a multi-billion parameter model. Not so remote, though – their server activity remains within EU jurisdiction.

Then, of course, there is the ‘peak privacy’ option – running your own LLM. That’s a lot easier than it sounds, thanks to easy-setup software like LM Studio or Ollama (both US-based projects, but run locally on your own machine). Install, download a model, and prompt away. While few (to no) people will have the hardware to run full-sized LLMs, small models are getting better and better, rivalling the biggies for everyday use.

Google’s Gemma 4 is a case in point, a new small model (you can get a sub-20gb version) achieving some really impressive benchmark scores. Multi-language support is one of its strengths, and believe me, it does more than a good enough job of worksheet authoring and lesson planning. And it comes with an extra ‘externalities’ bonus, too – the only energy it’s using it your laptop battery, rather than spinning up some red-hot servers on a remote farm somewhere.

That has to be a win-win – using open source releases from the industry leads, without getting trapped inside the matrix.

We may have little control over geopolitics. But there are always choices when it comes to our exposure to it in the tech we use. I’m working on a list of these swaps as part of my own digital hygiene plan, and hope to share much more of this in coming weeks!