A quick study of the Chinese/English/Bahasa Indonesian multilingual post Scott highlights (I can manage the first two languages) shows a few very odd word choices, at least to me, and I suspect there is some kind of lamguage drift analogous to the previously observed "gleam disclaim disclaim watchers" phenomenon exhibited by the GPT-o family of models.

Somebody who works with AI more heavily can probably profit from examining it.