Mini Guide to Prompting

Mini Guide to Prompting

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

25+ Hilarious Chat GPT Memes that You Cannot Miss in 2023

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

  1. You are a food critic writing for the Michelin Guide. Write a review of [pizza place].

  2. 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

  1. https://learnprompting.org

  2. https://www.promptingguide.ai

Did you find this article valuable?

Support Developer's Odyssey by becoming a sponsor. Any amount is appreciated!