Nice article!

You said: "Therefore, as they have ingested basically all the Internet, a place where people vent their inner racist for free, these models easily became racist too, or homophobic, and can even become hate-speech machines, making the interacting experience as a whole a very unpleasant one."

Question: what do you think of changing "these models easily became racist" to be something like "these models seemingly can be made racist"... models are a function of the corpus on which they're trained... in themselves, they're not racist or homophobic or age discriminatory.

The change would more directly acknowledge your point about the role that humans have in LLM development and use, as well...

Again, thanks for your article!