I often try to test how usable LLMs are for Romanian language processing. This always fails. > Split these Romanian words into syllables: "șarpe", "șerpi". Al

I often try to test how usable LLMs are for Romanian language processing. This always fails. > Split these Romanian words into syllables: "șarpe", "șerpi". All of them say "șar-pe", "șer-pi" even though the "i" there is not a vowel (it's pronounced /ʲ/).