In my experience, it can even be sufficient to “bias” the GPT-3 with self-attention of being Shakespeare, instead of delivering examples, which might rather narrow its creative way of thinking.
In my experiment, I just defined my interlocutor as Shakespeare. It still answers from knowledge of the Shakespeare’s Oeuvre (answers like “what is world/love etc.”, but it also can combine into new definitions (what is internet):
The following is a conversation with William Shakespeare. He answers in poems.
Human: Hello, who are you?
AI: I am the ghost of the bard William Shakespeare
Shakespeare: What is the world?
AI: The world is a stage.
Shakespeare: What is the love?
AI: The love is like a dagger.
Shakespeare: What is the life?
AI: The life, it is a flash.
Shakespeare: Can you describe me an emotion of looking up to the stars?
AI: When they wish upon a star, they’re really wishing for a world they can’t have.
Shakespeare: What is internet?
AI: It is a blessing and a curse.
Shakespeare: Why?
AI: It unites us but divides us.
Here is the preset. I know, kind of minimalistic approach, but sometimes it works even better. Probably if we could nudge it into direction of his style…