Prompting 101 with Jethalal – AI Seekhne Ka Desi Tareeka 😎

Table of contents
- ⚙️ Prompting Styles (Technical Formats)
- 🪄 TL;DR Table
- 🛠️ Prompting Techniques
- 1. Zero-shot Prompting – “Bina context ke kaam chalu”
- 2. Few-shot Prompting – “Thoda dikhake kaam karwana”
- 3. Chain-of-Thought (CoT) Prompting – “Soch samajh ke bolna”
- 4. Self-Consistency Prompting – “Akele decide karo”
- 5. Instruction Prompting – “Seedha bol bhai, kya chahiye?”
- 6. Direct Answer Prompting – “Seedha jawab do, haan ya naa.”
- 7. Persona-Based Prompting – “Act like someone”
- 8. Role-Based Prompting – “Ab tu engineer ban gaya”
- 🛍️ In Conclusion:
So you’ve heard everyone saying “write better prompts, bro” but you're just sitting there like Jethalal when Bhide talks about society rules—clueless but confidently wrong.
Don’t worry. Let’s make this easy by using our beloved Gada Electronics owner to explain the different types of AI prompts.
⚙️ Prompting Styles (Technical Formats)
🧠 These formats tell the model: “Boss, ye hi tareeka hai baat karne ka.”
1. 🦙 Alpaca Style
Very plain and simple. You use an Instruction + Input + Output structure.
pythonCopyEditprompt = """### Instruction:
Write a story about chai in 3 lines.
### Input:
None
### Response:
Ek garma garam chai, ek baarish bhari gali, aur ek broken heart – perfect kahani!
"""
Jethalal logic: “Seedha likho — kya karna hai, kya diya gaya hai, aur kya chahiye. Jaise school ka exam paper.”
2. 📋 INST Style (Instruction-tuned models like FLAN, LLaMA2)
Use tags like <<SYS>>
, ### Instruction
, ### Response
.
pythonCopyEditprompt = """
<<SYS>>
You are a helpful assistant who answers like Jethalal from TMKOC.
<</SYS>>
### Instruction:
Tell me how to make fafda-jalebi.
### Response:
Arre Daya! Fafda le, usko tel mein tal, aur jalebi ko syrup mein ghuma de! Bas ho gaya bhai!
"""
Jethalal logic: “System ke rules pehle bata do. Fir instruction. Fir jawab. Jaise shop ka SOP manual!”
3. 💬 ChatML Style (used by OpenAI Chat Models)
Structure like a conversation using role: user
, role: assistant
.
pythonCopyEditmessages = [
{"role": "system", "content": "You are Jethalal from TMKOC."},
{"role": "user", "content": "Why should I never skip breakfast?"},
{"role": "assistant", "content": "Arre bhai, bina nashta ke dimaag kaam nahi karta! Shop jaane ka mood bhi nahi banta!"}
]
response = openai.ChatCompletion.create(
model="gpt-4",
messages=messages
)
Jethalal logic: “Jaise Gokuldham mein conversation hoti hai — har role ke hisaab se jawab!”
🪄 TL;DR Table
Prompt Style | Format Elements | Model Examples |
Alpaca | Instruction / Input / Response | Alpaca, Stanford |
INST | <<SYS>> , ### Instruction | LLaMA 2, FLAN |
ChatML | role-based messages | OpenAI, Claude |
🛠️ Prompting Techniques
Let's explore various prompting techniques with examples.
1. Zero-shot Prompting – “Bina context ke kaam chalu”
👉 What it is: You ask the AI to do something without giving any example.
🧠 Example:
“Write a joke about Gujarati food.”
👴🏼 Jethalal logic:
Like when Popatlal proposes to a random girl without even knowing her name. Zero prep, zero clue.
2. Few-shot Prompting – “Thoda dikhake kaam karwana”
👉 What it is: You give a few examples before asking the AI to continue in the same style.
🧠 Example:
"Here are 2 funny tweets about society drama. Write a 3rd one."
👴🏼 Jethalal logic:
When he shows a few YouTube reviews of a product to impress a customer, hoping they’ll buy without question.
3. Chain-of-Thought (CoT) Prompting – “Soch samajh ke bolna”
👉 What it is: You tell the AI to think step-by-step before answering.
🧠 Example:
“Solve this math problem step by step: 128 x 23.”
👴🏼 Jethalal logic:
Like when he tries to explain to Bapuji why he didn’t go to the shop—starting from rain to traffic to chappal toot gaya. 🥲
4. Self-Consistency Prompting – “Akele decide karo”
👉 What it is: AI tries multiple reasoning paths and picks the most consistent answer.
🧠 Example:
“Try solving this riddle in a few different ways and give the best answer.”
👴🏼 Jethalal logic:
Like when Jetha calls Bagha, Natu Kaka, and Bapuji for advice—then does what makes the most sense (usually after getting scolded).
5. Instruction Prompting – “Seedha bol bhai, kya chahiye?”
👉 What it is: You give clear instructions on what you want.
🧠 Example:
“Write a 3-line poem about chai in a funny tone.”
👴🏼 Jethalal logic:
When Bapuji says, “Subah 7 baje uthna. Walk pe jaana. Nashta ready hona chahiye.” No confusion, only instructions.
6. Direct Answer Prompting – “Seedha jawab do, haan ya naa.”
👉 What it is: You ask for a straight answer—no stories, no drama.
🧠 Example:
“What’s the capital of Gujarat?”
👴🏼 Jethalal logic:
When Inspector Chalu Pandey asks: "Chori tune ki ya nahi?" And Jetha says: "Nahi!" End of story.
7. Persona-Based Prompting – “Act like someone”
👉 What it is: You ask the AI to answer as a specific personality or type of person.
🧠 Example:
“You’re a Bollywood gossip queen. Spill tea about the latest award show.”
👴🏼 Jethalal logic:
Like when he talks like Bapuji in front of customers to sound mature—but 2 mins later is dancing to “Aivayi Aivayi.”
8. Role-Based Prompting – “Ab tu engineer ban gaya”
👉 What it is: You assign a professional role to the AI.
🧠 Example:
“Act as a CA and explain how to file taxes for a small business.”
👴🏼 Jethalal logic:
When he tells Babita ji: "Main toh electronics expert hoon, madam."
Meanwhile Bagha is doing all the work. 😆
🛍️ In Conclusion:
Prompting isn’t rocket science—it’s just knowing how to ask properly. Even Jethalal can do it (okay, maybe with a little help from Taarak Mehta).
Next time you open ChatGPT, ask yourself:
“Am I prompting like a pro or like a Popatlal?”
Choose wisely.
#ChaiCode #ChaiAndCode #GenAI #Gratitude #LearningJourney #AIprompting #PromptEngineering #ChainOfThought #ZeroShotLearning #FewShotLearning #SelfConsistency #OpenAI #LLaMA #ChatGPT #ChatML #PromptStyles #Alpaca #INSTformat #PersonaBasedPrompting #RoleBasedPrompting #AITutorials #GPTtechniques #AIexamples #PromptWritingGuide #AIContentCreation
Subscribe to my newsletter
Read articles from Tushar Parmar directly inside your inbox. Subscribe to the newsletter, and don't miss out.
Written by
