How Large Language Models (like ChatGPT and Gemini) Work: A Plain-English Guide for Students

AI chatbots can explain photosynthesis, write code and plan your study week. It feels like magic, but the core idea is surprisingly simple. Understanding it helps you use these tools well and know when not to trust them.

The core idea: predict the next word#

A large language model (LLM) is a program trained to do one thing: given some text, predict what comes next.

If you read "The capital of Bihar is", you would predict "Patna". An LLM does the same, but it has practised on a huge amount of text, so it has learned patterns of grammar, facts, reasoning styles and even code.

To write a full answer, it predicts one piece, adds it to the text, and predicts again, over and over:

Question: What is photosynthesis?
Step 1 → "Photosynthesis"
Step 2 → "Photosynthesis is"
Step 3 → "Photosynthesis is the"
...       (until it predicts an "end" marker)

Tokens: how the model reads text#

Models don't see words the way we do. Text is split into tokens, small chunks that are often parts of words.

This is why models sometimes struggle with letter-counting puzzles ("how many r's in strawberry?"): they don't see individual letters, they see tokens.

Training: learning from examples#

Training happens in stages:

1. Pre-training: the model reads billions of sentences and repeatedly guesses the next token. Each wrong guess slightly adjusts its internal numbers (called parameters, often billions of them). Slowly the guesses get better.
2. Fine-tuning: it is then trained on examples of helpful conversations, so it learns to answer questions instead of just continuing text.
3. Feedback: people rate answers, and the model is tuned to prefer clear, safe and useful replies.

Context window: the model's short-term memory#

Everything in your conversation, your question plus the earlier messages, is fed to the model each time. The maximum amount it can consider at once is its context window. Very long chats may push early details out, which is why a bot can "forget" what you said at the start.

Why AI sometimes makes things up#

Because the model predicts likely text, it can produce an answer that sounds right but is wrong. This is called a hallucination. Common examples:

• Inventing a formula, a date or a quote that looks plausible.
• Giving a confident answer to a question that has no single answer.
• Making arithmetic slips in long calculations.

How to use AI to study smarter#

AI is best as a tutor that explains, not a machine that does your work.

1. Ask for explanations at your level. "Explain recursion to a Class 11 student with a real-life example."
2. Ask it to quiz you. "Give me 5 MCQs on the French Revolution, one at a time, and wait for my answer before showing the next."
3. Explain back. Write your own explanation and ask the AI to point out mistakes. Teaching is the fastest way to learn.
4. Get step-by-step help, then solve the next one alone.
5. Make a study plan. "I have 30 days and 2 hours a day for BPSC TRE Computer Science. Make a weekly plan with revision days."

Good prompts vs weak prompts#

Key terms at a glance#