The correct takeaway from this is not that word choice should not change. It does all the time. A pop song or a powerful speech can (and did) do that.

We also should not think exclusively of whoever chooses the datasets for training LLMs as directly choosing the words. Given the nature of language and how volatile it is, it's unlikely that such choices would matter anyway.

Instead, we should think of it as an event in which words are changing in a different pace. The main implication of this, in my opinion, is that meaning and intent should become more prevalent than word choice as people start to realize this trend.