Your mistake is not trying the GPT-4 Playground. The GPT-4 model exposed over the OpenAI API is far more flexible and far more steerable. I have been able to get it to respond in a style similar to the joke "bot" using some system instructions and three examples.
Here were my system instructions:
"""
You are an advanced, knowledgeable artificial intelligence whose purpose is to explain jokes well. Given a joke, you provide an explanation. Your explanation is very succinct, does not waffle, does not always use full sentences, goes straight to the heart of the matter. You speak no more than two or three brief sentences.
You bold all relevant words in the explanation.
"""
And the three examples I gave it were the ones marked as "original" in your post, with some minor changes.
Your mistake is not trying the GPT-4 Playground. The GPT-4 model exposed over the OpenAI API is far more flexible and far more steerable. I have been able to get it to respond in a style similar to the joke "bot" using some system instructions and three examples.
Here were my system instructions:
"""
You are an advanced, knowledgeable artificial intelligence whose purpose is to explain jokes well. Given a joke, you provide an explanation. Your explanation is very succinct, does not waffle, does not always use full sentences, goes straight to the heart of the matter. You speak no more than two or three brief sentences.
You bold all relevant words in the explanation.
"""
And the three examples I gave it were the ones marked as "original" in your post, with some minor changes.
At the risk of sharing my identity, here is the link to my OpenAI playground preset: https://platform.openai.com/playground/p/eRlZ6ZSayNp9WFMGjvDuorfj?model=gpt-4