Sidenote: from the pov of descriptive linguistics LLMs _seem_ amazing: a fairly true representation of the consensus we call our language.
But, this gets more complicated in practice, as the dataset will be always biased.
So in a sense the opposite is true: LLMs are prescriptivist as they reflect the biases of the people involved in their creation, thus introducing a "standard" version of the language(s), just not in a completely deliberate way.
Sidethought to sidenote: I'm now imagining the Académie Française engaging in huge state-funded black-budget operations to infiltrate language models to finally enforce the correct French language standards once and for all.
> to finally enforce the correct French language standards once and for all.
*the one and only true and perfect French language standard
When you look at the Académie Française's history and origins, it’s quite jarring.
I will never forget Mézeray's preference for « l’ancienne orthographe, qui distingue les gens de Lettres d’avec les Ignorants et les simples femmes » ("the traditional spelling that sets apart the educated from the ignorant and women", women as a whole being described as ordinary simpletons).
Another funny thing is how they aren’t linguists. Just writers. Not language experts. Writers. Experts at writing stuff liked by the people who pick "Académiciens".
But, this gets more complicated in practice, as the dataset will be always biased.
So in a sense the opposite is true: LLMs are prescriptivist as they reflect the biases of the people involved in their creation, thus introducing a "standard" version of the language(s), just not in a completely deliberate way.