You are on page 1of 8

6 TIPS FOR

Prompt Engineering

Saurabh Naik
Use Delimiters
01 This method is used to
tell LLM about distinct
part of user input from
total prompt.
Also better to avoid
prompt injection issue
SPECIFY
02 OUTPUT
FORMAT
This technique is
useful when we want
output in some other
form than direct text
like bullet points, json
format.
Check whether
conditions are
satisfied
03 When the user does not
provide ideal input which
we assumed that they
will provide we can use
this strategy to suggest
them to provide correct
input.
This method can be
assumed as error
handling in prompt
engineering
Few shots
prompting
LLMs generally perform

04
better results when we
provide them some
examples in the prompt.
If we provide them one
example it is called one
shot prompting.
Experiment with random
number of examples for
your prompt and check
output
Specify steps
05 to complex
task
To do reduce
hallucinated results
from LLMs, it is a good
idea to provide steps to
solve a complex task in
order provide accurate
results
06 Let model figure
out
Instruct the LLM to come
up with it's own solution
and then compare it with
the solution provided by
the user.
For example: consider
students solution for a
mathematical problem.
FOLLOW ME
For more such Data
Science Content

Saurabh Naik

You might also like