Slightly out of left field - I'm using it to generate Chinese conversations for foreign language students, paired with TTS voices and 3D avatars, streaming live on Twitch.

https://www.twitch.tv/polyvoxlive

It's not 100% flawless, but for this task, it is exceptionally good (in particular, it makes fewer mistakes than the davinci models).