I tried the same Ai and asked it to provide a list of 20 things, it only gave me 5. I asked for the rest and it also apologized and then provided the rest. It’s weird that it stumbles at first but is able to see it’s error and fix it. I wonder if it’s a thing that it ‘learned’ from the data set. People not correctly answering prompts the first time.
Might be an intentional limitation to avoid issues like the “buffalo” incident with GPT3 (it would start leaking information it shouldn’t after repeating a word too many times).
I personally don’t think a large section of the population meets the requirement for general intelligence so I think it’s a bit rich to expect the AI to do it as well.
It’s weird though because 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.
I’ve never been able to get an AI to explain its logic though which is a shame. I’m sure it would be useful to know why they come up with the answers they do.
I’ve never been able to get an AI to explain its logic though which is a shame. I’m sure it would be useful to know why they come up with the answers they do.
you and AI researchers both. it’s probably a trillion-dollar problem at this point
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.”
AI will make things better
Hahaha, I like that having it re-read the question fixed the issue…
I tried the same Ai and asked it to provide a list of 20 things, it only gave me 5. I asked for the rest and it also apologized and then provided the rest. It’s weird that it stumbles at first but is able to see it’s error and fix it. I wonder if it’s a thing that it ‘learned’ from the data set. People not correctly answering prompts the first time.
Something else i also encounter with gpt4 a lot is asking “why did you do x or y” as a general curiosity of learning how it handles the task.
Almost every time it apologizes and does a fully redo avoiding x or y
Might be an intentional limitation to avoid issues like the “buffalo” incident with GPT3 (it would start leaking information it shouldn’t after repeating a word too many times).
If only that worked on humans!
I personally don’t think a large section of the population meets the requirement for general intelligence so I think it’s a bit rich to expect the AI to do it as well.
We all know the first black man in space was George Santos.
I still want to know what the fucking fuck triggered “possible self harm” in your first question.
It’s weird though because 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.
I’ve never been able to get an AI to explain its logic though which is a shame. I’m sure it would be useful to know why they come up with the answers they do.
you and AI researchers both. it’s probably a trillion-dollar problem at this point
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.”
Just early days of AI fun. Neo probably disconnected from his goo bath or sum idk.
Normally it ends the conversation at this point and refuses to answer any thing else, disabling the text box. At least it let you try again!