The model says it has been trained in a variety of languages, but its current version is english. Its translations are bad, probably because it uses a translation model on top of its output. Sometimes it is obvious it is translating english articles from the mistakes it makes
I don't know what it uses to output non-English text, but if you address it in a different language, it'll quite happily speak that back to you. And I'd say it understands it just fine, not really any worse than English - although it sometimes gets the word forms wrong in inflective languages.
Yup, doesn't rhyme for me, either, if I ask for rhymes or poetry in Russian (but I don't think it'd rhyme even in English).
I think that makes sense if the supermajority of the training data was in English - it'd make it heavily biased towards that by default. So even when it speaks another language, it still "thinks" of the words in terms of questions like "how it sounds". I suspect it might be possible to craft a sufficiently detailed prompt to work around that.
The original GPT-3 also did bad translations, just from scraps of foreign languages in the dataset. They seem to be worse though, so perhaps you're right that Chat-GPT is using another model the way it uses Internet searches
