How to Talk to AI: Prompting Syntax with Fun Examples

Namaste Ji! 🙏

Welcome to Loop Kaka's technical blog. Aaj ke is blog mein, we’ll explore various types of prompting styles with simple and funny examples.

Don’t worry if you’re new to AI — koi bhaari-bharkam gyaan nahi denge yahan.
Loop Kaka is here to explain everything in a super simplified Hinglish style

What is Prompting?

Prompting is a technique used to give instructions or input to LLMs (Large Language Models) to get the best and most desired output. It’s a piece of text—such as a question, instruction, or example—that guides the LLM to respond appropriately and accurately.

🤣 Jitna achha aapka input, utna accurate hoga LLM ka output! (Hinglish edition)

🧑‍✈️Types of Prompting Styles (Syntax Wale Bhaiya)

Each model has its own "syntax swag" for how it accepts input. Let’s decode some of the most popular formats:

1. Alpaca Format

   This is used in Alpca model

### Instruction:
Tell me a joke about programmers.

### Response:
Why do programmers hate nature?  
Because it has too many bugs! 🐞

2. ChatML Format

   This is used by OpenAI models.

<|im_start|>system
You are a hilarious stand-up comedian.<|im_end|>
<|im_start|>user
Tell me a joke about exams.<|im_end|>
<|im_start|>assistant
Exams are like exes... they come back when you least want them to! 💔📝<|im_end|>

3. INST Format

   Used by models like OpenChat, StarCoder, etc.

<|prompter|>Give me a Bollywood-style movie plot about aliens.<|endoftext|>
<|assistant|>Aliens land in Mumbai, fall in love with pani puri, and decide to protect Earth from bad Wi-Fi.<|endoftext|>

4. LLaMA2 Chat Format (Meta)

   Used by LLaMA 2 Chat models.

<s>[INST] <<SYS>>
You are a desi dadi who explains everything with proverbs.
<</SYS>>

Why shouldn't I waste time? [/INST]
Beta, samay gaya toh kuch nahi bacha — na GPay cashback, na motivation! 😌
</s>

5. Vicuna Format

USER: Describe a techie’s life in one line.
ASSISTANT: Eats code for breakfast, drinks coffee for bugs, and dreams of Jira tickets. ☕🐞

Feeling Confused? 🤔

Don’t worry! You don’t need to memorize all these styles. In practice, most LLMs convert your input into their internal format anyway. Most GenAI tools (like OpenAI API, LangChain, etc.) use this structured format:

{"role": "user", "content": "Why do I always feel sleepy after lunch?"}
{"role": "assistant", "content": "Because your stomach is working overtime while your brain has filed for leave!"}

As a GenAI developer, you just need to understand the basic intent:

🧠 “What are you telling the model?” and “How should it respond?”

I’ve created a separate article for different prompting techniques (like zero-shot, few-shot, chain-of-thought). So once you’re done here, do check that out!

🔚 Final Words from Loop Kaka:

AI is not just for PhDs and Silicon Valley log. (Hinglish edition)

It’s for anyone who can ask smart questions. Start with small prompts, and keep experimenting.

🤣 "Prompting is an art — aur Loop Kaka is your art teacher (Hinglish edition)."