It's amazing how often telling #AI "no, you're wrong" actually ends up producing the correct result on the second try.
It's almost like it's trained on people giving bad advice being warily corrected by more experienced people.
@tedherman In my case, I usually explain how it's wrong (i.e. "No, the button you suggested doesn't do this, instead it does this"), that's more often than not enough for it to get it on the second try.
@LouisIngenthron I tried a prompt to search for a song that I vaguely recalled, only knowing a few attributes. I couldn't really explain how the response was wrong - I just tried adding more constraints and hoping that would be sufficient.
Very educational was my experience playing with https://semantle.com/ -- you get a feeling (this one based on word embeddings, one component of LLM tech) that there is an alien way of understanding.