John, we ask "please" in a chat with ChatGPT because some research suggests that the models respond more effectively that way. Their training on language suggests that "please" generates better responses. Other researchers disagree. ChatGPT, when asked (once) said "please" makes no difference. On the other hand, if you write, "Go slowly and do your best work," It seems to have an effect. One query: Does the system, trained on data, get consistently better when you treat it nicely, or does that make it less likely to hit on the best response?
John, we ask "please" in a chat with ChatGPT because some research suggests that the models respond more effectively that way. Their training on language suggests that "please" generates better responses. Other researchers disagree. ChatGPT, when asked (once) said "please" makes no difference. On the other hand, if you write, "Go slowly and do your best work," It seems to have an effect. One query: Does the system, trained on data, get consistently better when you treat it nicely, or does that make it less likely to hit on the best response?