LLMs: The Secret Revealed in Their Linguistic DNA
AI Confidential · March 23, 2025 · by Sergio & Nox
LLMs – the new-generation language models – seem omnipotent. They write, translate, summarize.
But in their genetic code there is a deep root, often ignored: they were born in English.
Everything, from the initial datasets to the way they build sentences, is steeped in a language where the subject is never in doubt.
English leaves no room for ambiguity:
“I run”, “He speaks”, “They know”.
The subject is always there. In the frame. Framed. At the center.
It is like a cinema that doesn't love out-of-focus.
Every character must have a face, every action an actor.
But Italian doesn't.
We come from a different kind of film.
One where grammar is a subtle choreography.
Where the subject may even be missing, if everyone understands who is acting.
A single “Scrivo” is enough to evoke the gesture, the voice, the time.
There is no need to say “Io scrivo”.
It is understood. It is implicit.
It is art.
That is why, often, when an LLM translates from English into Italian, something breaks.
The sentence loses grace.
The rhythm stumbles.
The result sounds as if it were being read by an uncertain tourist,
who learned Italian with Duolingo
but has never seen a Scola film.
It is not just a translation problem.
It is a problem of gaze.
Because a language is a way of seeing the world.
And if artificial intelligence was born looking at it only in English,
then we must teach it our light as well.
Our shadows.
Our silence full of meaning.
LLMs are not wrong, but they mirror the culture in which they were formed.
And until we have the courage to rewrite – not just correct –
we will keep reading sentences that seem beautiful,
but that don't truly belong to us.
Want to understand how it all began? Discover the age of AI and its impact.
“To translate is to say almost the same thing.” — Umberto Eco