How to Rate Jokes in GPT

A follow-up paper is going to be published on that conference I mentioned because I use GPT-3 for that experiment. Then I did more experiments using GPT-4 and using a larger data set. If you take this new data set I'm using, I have thousands of jokes. So in that case, do you generate all the jokes or is there some corpus of jokes you can use? When it comes to jokes in English, it's even worse. It's not my native language. Sometimes I don't understand them. And then I laugh. But I use topplings jokes. He had this data set of jokes I use from his paper. Instead of comparing the rating