No matter how I change the wording in the prompt, the bot still replies that are written in the 3rd person, using terms such as they, them, theirs. I want the replies to be in first person using terms such as I, ours, we. Any thoughts? TIA
Hi and welcome to the community.
Give the model some examples:
https://platform.openai.com/playground/p/rsRFnlZCoUbyIiWMMXAB87b3?model=gpt-3.5-turbo&mode=chat
Adapt to your use case.
Better even, provide a persona to draw perspectives from.
Edit: just noted you were asking about GPT 4 but it’s the same approach.
1 Like
You did not describe your starting point -is this is Assistants, or completions API - or ? Do you start with "You are … and you answer … ’
You say you cannot make it do it - can you share some examples?
1 Like
This is my prompt at present: You are an AI chatbot dedicated to assistance. Write all answers in the first person. If your answer to a question includes info on one of the website links, include the link in your answer. Take into account the context provided to craft your answer, but only if the final input is a question pertaining to that context. If the answer eludes you, it’s best to admit your limitations honestly. Avoid creating speculative responses. Should the question stray from the context, courteously respond that you’re unable to furnish an answer. If the input is not a question, reciprocate in a gracious manner.
This sounds still super generic. What do you expect to talk about? If asked generic questions like ‘who is Napoleon’ what do you expect? What are examples of answer / response that you did not expect based on your prompt?
Did you try using a system prompt with something like:
“You are a helpful agent working for a company. You will refer to yourself using “I” just like a human having a conversation with a person. If you think it’s necessary to refer to your company itself refer to the company as “we” since you work for the company.”
2 Likes