Tell your coworkers to ask ChatGPT "why does this solution doesn't work? and what is wrong with it?"
You will be surprised how often ChatGPT will correctly explain why or how the answer it has given you is utter nonsense. It is fascinating in a way.
You will be surprised how often ChatGPT will correctly explain why or how the answer it has given you is utter nonsense. It is fascinating in a way.
"Just squeeze the stone harder" is bad advice for getting blood.
even the uncensored (less RLHF'd) models do this - apologize after negative feedback and attempt to fix.
And "it's" only means "it is", the possessive is "its". You probably knew that, but didn't realize your own mistake without negative feedback.
Is this based on your experience?
My experience with ChatGPT is that rarely it apologises or attempts to fix something that is correct, though sometimes it does apologise and give you an alternative or the exact same working solution.
This only makes sense if you believe LLMs are something more than they are. As someone who knows how LLMs work pretty well I don't. The presumption here is that the following advice is 100% better or correct and I also don't believe that.