Last year they replaced all the recordings of native speakers with ML-generated recordings, in both Russian and Ukrainian (probably other languages too but those are the two I have). The ML-generated recordings are terrible, for example they can't deal with the ambiguity betweeen vse and vsyo (written identically in Russian) so they always say vsyo. They'll sometimes randomly say the names of individual letters instead of reading the word, particularly the hard and soft signs. One recording is for a sentence with the word "tochka" (period, as at the end of the sentence) and instead of reading "tochka" the recording just has a silence there.
I've reported these issues hundreds of times since they added the ML recordings and none of them have been fixed.
But like you I keep using it just to get that little daily exposure to the language. I suspect it's useless for actually learning a new language, but it's maybe just barely good enough to keep up a language you already know.
The ambiguity goes away if you write them as "все" and "всё". Diacritics, accents, umlauts and cedillas are useful; if you omit them you do a disservice to the reader.
Maybe so, but omitting them on ё is standard practice in Russian. A language learner who doesn't learn to deal with the ambiguity of the omitted diacritics will struggle to read real-world texts.
I've reported these issues hundreds of times since they added the ML recordings and none of them have been fixed.
But like you I keep using it just to get that little daily exposure to the language. I suspect it's useless for actually learning a new language, but it's maybe just barely good enough to keep up a language you already know.