10 AI topics, one page.

01

AI ante enti?

Idea

Getting computers to do jobs that normally need human intelligence: understanding language, recognising pictures, making decisions.

Example

Your phone gallery finds every photo of your friend when you search their name. Nobody tagged them; the AI learned to recognise faces.

Habit

Treat AI as a smart helper, not magic. It learns patterns from data, so it can be very useful and still be wrong.

02

AI model ante?

Idea

The engine that actually gives the answer. It is trained on data, and what it learned is saved as billions of numbers called parameters (or weights).

Example

You ask a doubt on WhatsApp. WhatsApp is just the window; your friend gives the answer. In the same way, the AI app is the chat window and the model behind it gives the reply.

Habit

Pick the model that fits the job. Small models are fast and cheap for simple work; big models are slower and cost more, so save them for hard thinking.

Go deeper: Complete AI Model Playbook

03

LLM ante?

Idea

Large Language Model: a model trained on a huge amount of text. It understands your text and writes new text by predicting the next small piece of a word, again and again.

Example

Writing an email, summarising a long message, translating, explaining a topic simply. The chat AI apps you use run on LLMs.

Habit

Use it for language work: drafts, summaries, explanations. It writes what sounds right, so check any fact it states before you use it.

04

Prompt ante?

Idea

The instruction you give the AI. A good prompt has four parts: role, task, context and format.

Example

"Write a leave mail" can come back with a fever you don't have. Add the context: "You are an office assistant. Write a leave mail to my manager for Friday and Monday, my cousin's wedding in Vizag. Five lines, polite."

Habit

Never skip the context. Without it the AI guesses and writes for everyone, not for you.

Ready to copy: Ready-Made AI Prompts · RCTRE Prompt Templates

05

Prompt life cycle

Idea

What happens in the few seconds after you press Enter: your prompt goes to a server, is cut into tokens, turned into numbers, and the model builds the reply one token at a time.

Example

You ask "Why is the sky blue?" The reply appears word by word on your screen because the model is still writing it. That is called streaming.

Habit

If a reply starts going the wrong way, stop it and rewrite your prompt. Every next word builds on the words before it, so a bad start rarely fixes itself.

06

System prompt

Idea

Instructions the model receives before your message, telling it how to behave. The app or the developer writes them; you usually don't see them.

Example

A shop's support chatbot is told up front: "Answer only about our products, be polite, reply in short steps." Every customer chat follows that.

Habit

If you build with AI, keep the system prompt short and general. It goes with every single request, and a narrow rule there affects every answer.

07

Custom instructions

Idea

Your own standing preferences, saved once in the AI app's settings, so you don't repeat them in every chat. It is your personal system prompt.

Example

At the tea stall you said "sugar takkuva" once; now "one tea" is enough. Same here: "Explain simply, with everyday examples" saved once applies to every chat.

Habit

Write only your general style. A rule like "answer in two lines" also cuts your resume and reports to two lines, because it applies to every chat.

08

Temperature

Idea

A setting that controls how random the AI's word choices are. Low temperature: almost the same answer every time. High temperature: more variety and creative answers. (Not a fever, Bhattu garu!)

Example

Ask amma "what's for tiffin?" Most days idli, some days dosa, once in a while upma. Ask the AI the same question twice and you can get two different answers the same way.

Habit

For ideas and captions, ask again to get new options. For a fact or a number, ask the same question twice: if the two answers differ, it may be a guess, so check it yourself.

09

Hallucination

Idea

When the AI says something false with full confidence. It is not lying; it predicts words that sound correct, and when it doesn't know, it fills the gap instead of saying "I don't know".

Example

A lawyer in the US submitted court cases that an AI chat had given him. The cases did not exist, and the court fined him.

Habit

For names, numbers, links, medicine, law and money, never trust the answer directly. Ask for the source, then open that link yourself: AI can make up sources too.

10

Tokens

Idea

The small pieces the AI cuts text into before reading it. A token can be a whole word or part of one. Usage limits and costs are counted in tokens, not messages.

Example

Your mobile data pack counts MBs, not videos. AI counts tokens the same way. And each time you send a message, the whole old chat goes to the model again, so even "ok" in a long chat costs a lot of tokens.

Habit

New topic, new chat. Your limit lasts much longer.

More habits: Cut Token Usage: 360° Guide