And unless I’m quite mistaken, any specific correction one might possibly contribute to the LLM project in question (i.e. software hallucination), is generally-speaking, roundly & enthusiastically embraced and even celebrated by the LLM, then immediately and completely ignored.

I.e., they’re not programmed to listen to our feedback in a meaningful, educational way, only to keep munching on the databases their doggie-daddies have sicced them upon.

EDIT: that cynicism / critique aside, ChatGPT in particular has been hugely useful in my language-learning, and there’s no question to me that it’s improved a lot, just across the last few months. FWIW

https://suno.com/song/f5f887be-c5b3-4a59-ae7f-21e8d1be5f80

I set it to play in the background and was thinking: ‘this is very good for AI, and a decent (but not very interesting) song,’ and then came back and took a closer look at the screen. Holy hell, my mind instantly asploded.

Matching the vocal so well to the text and having it sound so nuanced and well-sung is… almost terrifying. oO