I was reading some chatGPT programming workflow Twitter thread the other day and the guy was unusually polite with the thing. Like, does saying "please" or "thank you" actually improve its performance or is it just tech roleplay?
Yes, in the early days of ChatGPT I had noticed that polite conversation leads to less adversarial responses. In retrospect it does seem logical that a language model would choose more amenable responses and be more willing to help when spoken nicely to. I do not know if that behavior has been programmed out of it by now, I simply talk nicely to it with please and thank you out of habit at this point.
The person who started the thread said "I don't want that." Explaining that other people do want that is a perfectly valid response. If you would like to further elaborate on why you think sugar is the right analogy here that would be very welcome too of course.
I was reading some chatGPT programming workflow Twitter thread the other day and the guy was unusually polite with the thing. Like, does saying "please" or "thank you" actually improve its performance or is it just tech roleplay?