Google

The "name:content" format is effective for structuring conversational prompts. For instance, a dialogue between two speakers, like John and David, can be represented by prefixing each line of dialogue with the speaker's name. This format allows the model to understand the conversational context and continue the dialogue appropriately.

Example of a Name:Content Prompt for Conversation

A developer is creating a prompt to make a language model act as a helpful support agent in a conversation. The goal is to have the model generate the next response for the 'Agent'. Which of the following prompt structures is most likely to produce the desired conversational continuation?

You want a language model to generate a response from the perspective of a character named 'Dr. Aris Thorne,' a skeptical scientist, who is reacting to a statement made by 'Jenna,' an enthusiastic explorer. Jenna has just said, 'I'm certain I saw a living dinosaur in the jungle!'

Write a complete, well-structured prompt that uses the `name:content` format to instruct the model to generate Dr. Thorne's skeptical reply. Your prompt should include Jenna's statement and set up the model to generate Dr. Thorne's line.

Crafting a Conversational Prompt

A developer is building a chatbot to simulate a conversation between a 'Customer' and a 'Support Agent'. They find the model's responses are inconsistent and do not reliably follow the conversational roles. Analyze the following prompt and explain the likely structural reason for the model's poor performance. Then, suggest a specific revision to the prompt that would more clearly guide the model.

Learn Before

Related