If you’re even little aware of what’s going on in the world its nothing but what i would like to call the AI revolution with new and better models releasing everyday with outperforming results which includes OpenAI’s GPT series or Google’s Gemini series or Anthrophic’s Claude and last but not least Chinese revolutionaries Deepseek r series
Today we are not going to be analyzing which is the best or which is the worst but we will be finding ways to utilize them with the best potential .One way is called prompt engineering but i wouldn’t use the word engineering because I like to see it as a subskill ( a skill that is part of and necessary to another more complex skill ) and not a independent skillset.
a more appropriate name would be just prompting
Do i really need it
If there is a way to prove prompting is needed for better communication with large language models is by seeing a few examples and keep in mind all fancy AI tools essentially are LLMs.
Bad prompt : Give me roadmap for studying Web Development
very vague
doesn’t present any details
doesn’t fully utilize the capabilities
Good prompt : Act as a coding tutor that creates study plans to help people learn to code. You will be provided with the goal of the student, their time commitment, and resource preference. You will create a study plan with timelines and links to resources. Only include relevant resources because time is limited. My first request - "I want to learn web development. but I do not know how to code. I can study 5 hours per week and only video resources. Create a study plan for me.
gives a role to the LLM
time and resource type specified
has more than 1 thing to keep in mind
Which prompt do you think will perform better ? obviously the latter one
you guys can try it out yourselves as a fun activity and comment the responses you guys have got 😊
Types of prompting
Zero shot prompting
Zero shot chain of thought
Condition prompting
one shot & Few shot prompting
Role prompting
Chain of thought prompting
Personal Credit prompting
Zero shot prompting
This is when you completely depend on the LLM pre trained data and previous existing data and no new data will be provided to the LLM about the output needed or no task specific data wont be provided.
this is the type of prompting people must use from day to day life and it is quite common
for example :
without giving examples the LLM has to work with the data and figure out what the response should be
this is used when you care about efficiency and speed more than the result and keeping in mind of the resource limitations
Zero shot chain of thinking :
its the same as zero shot but adding a subtle “Lets think step by step” after the prompt helps the model immensely , lets see a short example
Condition Prompting
it is a method of applying continuous pressure to the standard prompting technique to meet the specified conditions mentioned by us
Goal : Push the model limits for desired results
for example :
Give me more example
Make it more human
Make the response more funny or serious
One shot and few shot prompting
This is when you give a set number of examples along with the prompt to make the LLM understand better about the question and use it to provide better responses
when you give one example to the model then its one shot prompting and when you provide more than 1 example then you can call it few shot prompting 👍👌
not so complex is it ??!?!?!?!
now we will go to the actual tedious type of prompting techniques
Role Prompting
you create a fictional role or character and make the model to role play as the imaginary character to provide us with perfect or desired results and responses.
this can sometimes be used when the model doesn’t give responses to a particular solution but when assigned a role or character it may try to give a proper response
but the main purpose of using role prompting is to get more authentic results from more nichy questions like the example we saw for creating a roadmap for learning a particular skill
like
You are a food critic writing for the Michelin Guide. Write a review of [pizza place].
You are a communications specialist. Draft an email to your client advising them about a delay in the delivery schedule due to logistical problems.
Chain of thought prompting
In this way of prompting , we will provide the model with the question to be solved along with a example of that particular problem and how to go about the problem and solve it a step by step fashion.
this increases their reasoning capabilities and helps them achieve some sort of efficiency and instead of looking directly for a solution , they solve the problem step by step
chain of thought (CoT) prompting is said or proved to bring a good amount of improvement to the model accuracy and quality of response
Personal Credit Prompting
this is a very special kind of prompting technique which can benefit everyone , so in this prompting technique
you give the model a role
mention how the result or response should be
give context of what’s going on
what’s the goal of this prompt
any particular constraints / conditions
a pretty good example of personal credit prompting is
You are an experienced UI/UX designer specializing in SaaS applications. Provide a step-by-step guide in bullet points on how to improve user onboarding for a productivity app. The goal is to enhance retention within the first 7 days. Ensure the solutions are cost-effective and can be implemented within a month
now what you do is
take the response / received output and ask the model to criticize it
how the model would prefer
rewrite / reiterate few of the steps
and be able to convince his results are better
this is an example of how it criticized his own results and made it better and honestly try it out guys , it brings surprising and in-depth results
You reached the end of this short mini guide on prompting , let me know if I can improve on anything and i am hoping to write 3 more blogs this month so stay tuned ( more about ai tools , real life use cases )
Examples were inspired from