I used the transformers.js sketch with the chatbot model to experiment with prompting. It was very straightforward to change the content of the chat history to change the response and character of the chatbot. I then tried the Replicate API through the proxy server and played around with the prompting for this API as well.

I started by just asking the AI to answer only in haikus. In the first example, I only prompted the AI with “You may only answer in haikus” and in the second example, I said “You are a poet. You may only answer in haikus.” It is surprising how small changes such as this affects the AI’s response to the question.

I then experimented with prompting the AI to be different personas. This is the difference between a DJ and a country artist. There is a clear difference in slang used between these two personas, which makes me wonder where these generalizations come from. This prompting definitely works well, as it is obvious to me which persona is which, but this perpetuates stereotypes of each of these things.

Reflection

The idea of any creative work being a result of everything which has come before it is not a new idea to me. But when it comes to AI, where it doesn’t seem to be capable of generation but instead just collaging related materials, I think there’s the sense that AI’s work cannot stand on its own. When integrated into my creative practice, I use AI has a helper, but I still feel a need to be the primary creator of a work, leaving my touch on my work even though I know that I myself am only a product of everything that came before me. So interestingly authorship and originality is all just up in the air now, especially with the increasing use of AI for creative practices.