they were able to point out they got to absurdity to its comment and it did agree. No it’s not just algorithmic phrase matching, there is an actual “thought process” going on.
Or it just knows to say those words when someone says “are you sure?” or something similar.
But then it provided the correct answer so it’s not just a rote response. If it was it would say no I am not sure, but then it wouldn’t be able to provide the response.
You could test it on a correct answer. Ask a question, see if it gives a correct answer, then ask “are you sure?” to see what kind of response it gives. My guess is that you won’t get an answer like “yes, I’m sure, that was the correct answer.”
Or it just knows to say those words when someone says “are you sure?” or something similar.
But then it provided the correct answer so it’s not just a rote response. If it was it would say no I am not sure, but then it wouldn’t be able to provide the response.
My guess is, when they get negative feedback they throw a bit more computing power into your instance for the second reply.
You could test it on a correct answer. Ask a question, see if it gives a correct answer, then ask “are you sure?” to see what kind of response it gives. My guess is that you won’t get an answer like “yes, I’m sure, that was the correct answer.”