Getting Started with Prompting

Modified on Tue, 4 Feb at 1:57 PM

Chat uses AI prompting to generate outputs from your content inputs.

Prompt Engineering — An Iterative Process

When entering a prompt, think of it like engaging in a dialogue. It's a process where you present an input, receive a corresponding output and then craft your subsequent input based on that output.

Once you receive your output, you need to evaluate it against certain criteria:

Quality: Is it sensible, specific and interesting?
Safety: Is it free of bias, hate speech, violence, etc.?
Groundedness: Is it supported by authoritative sources?
Informativeness: Is it a general, objective consensus on a certain topic?

Then, based on your evaluation, you can refine your prompt to:

Clarify or enhance the role, output or instructions.
Alter the wording or format of the prompt.
Add more context or detail.
Give feedback regarding parts to maintain or remove.

Modify to resolve issues identified while evaluating the output.

Prompt Structure

A good prompt structure contains three essential elements:

Audience role (LLM persona)
Output
Instructions

This example identifies each element in the prompt:

Structure:	Prompt:
Audience role	Act as Training Manager.
Output	Write the opening paragraph of an announcement to the organization regarding a new program on prompt engineering for all employees.
Instructions	The announcement should be short, engaging, using professional language and be well-structured.

This is a very basic starting guide to get you going on prompting.