This article shows a reason why I often joke that I refuse to use LLM-powered systems to avoid being “tainted”.
Among humans, language has harmless local feedback loops. First you don’t know what “on fleek” means, and you start using it more or less voluntarily, to the visible displeasure of your teenage kid.
But the feedback loop with LLMs is global and one-way since individual context windows are small and temporary (or permanent and, I assume, costly).
And for the same reasons I shouldn’t use “on fleek” seriously, I do not want my language to be converging with LLM systems. It simply isn’t the crowd I want to be associated with.
And while I (still) have a choice, the people the LLMs used the language style from are rightfully upset, just like the artists whose style have permeated most AI-generated images.