PROMPT ENGINEERING
BETTER PROMPTS FOR BETTER RESULTS
PROMPT ENGINEERING
It Is all about writing prompts to guide the AI
models to generate the most useful and precise
responses.
But how do we evaluate the AI model’s response?
1
EVALUATING THE MODEL’S RESPONSE
1. Is is clear and consice?
2. Does it address my needs?
3. Is it in the desired scope, format and tone?
4. Does it match my level of experience and
understanding of the topic?
2
PROMPT ENGINEERING
TECHNIQUES
3
Refine the prompt by adding a The role define the character that AI
model plays, and the audience define
role and an audience who the AI talking to
We give the AI model the ability to
narrow down the specific context of the Encouraging Clarifying questions
query letting it decides what to focus on
4
We break down complex questions or
task to small and easy to understand Decomposition
parts
We define the format, the constraints,
Defining the output format the level of language and the
convection of the response
5
WHY THIS TECHNIQUES WORKS?
The model tailors its word choice, in order to match Most of the time the initial prompt does not give the
the level of understanding and interests of the model enough context for an accurate response
audience you pick
The model becomes more efficient with the tokens The model generate the response according to the
it uses, spreading out processing power on more desired format (email, bullet list, text...), following the
specific requests. constraints (length of email, resources to
inlcude/exlude...) in the desired level of language
and tone (sarcastic, professional, emphatic...)
6
The output of the previous prompt
becomes the input for the next Prompt Chaining
AI models identifies areas of
Self-Critique improvements and flaws in their own
responses
7
Use popular frameworks/templates like:
Problem-Solving-Result, Before-After-
Creative
Bridge, STAR, AIDA, Five Ws and one H.... Frameworks
Reverse We start with the goal objective in mind and ask
the model to generate a prompt that would
Engineering generate the model response
8
THIS TECHNIQUES ARE GREAT WHEN
The model need the previous responses to When working with clear objectives and want your
generate more detailed final response prompt to deliver precise and actionable results
There are potential improvements and flaws in the You want a templates structure, that format the
model’s response model’s response in a specif method
9
THANK YOU FOR YOUR
ATTENTION