Comment by simion314
2 years ago
I see they use a lot of "do not do X" , for me this does not work. For example I ask it to write a few paragraphs about a topic and it always generates a conclusion, it is almost impossible to craft a prompt to make it not generate the conclusion, or feed it the text again and asking it to remove the conclusion. It is like the training is so strong in it that it always has to put a conclusion like paragraph at the end...
so is frustrating when it acknowledges it failed to follow your promt and not add the conclusion, it claims it will remove it but rewrites the text with the conclusion again.
I am surprise that a "do not" in the prompt actually works.
They’ve most likely used RLHF to make sure it follows these instructions… and we know they’ve also used RLHF to make it produce essays with conclusions.
Same technique, but it’s benefiting them and annoying the rest of us.