Meet the Models: ChatGPT, Claude & Gemini

How LLMs work, in plain English

Here is the surprising truth at the heart of every AI chatbot you have ever used: underneath all the magic, a large language model (LLM) is doing one thing extraordinarily well. It predicts the next word. That's it. You give it some text, and it asks itself, "Given everything I have read across billions of pages, what word is most likely to come next?" Then it adds that word, looks at the new, slightly longer text, and predicts again. Word by word, it builds a complete answer.

Think of it like the autocomplete on your phone, but trained on a library the size of the internet and far, far better at understanding context. When you text "I'm running a little..." your phone suggests "late." An LLM does the same thing, except it can carry that prediction across whole paragraphs, hold a tone, follow instructions, and weave in facts it absorbed during training. The fluency feels like understanding, and for practical purposes it often behaves like it. But the engine is prediction, not a database lookup and not conscious thought.

Why does this matter to you as a prompt writer? Because if the model is constantly guessing the most likely continuation of your words, then the words you choose steer everything. A clear, well-framed prompt nudges those predictions toward exactly what you want. A vague one leaves the model guessing in the dark. The whole craft of prompt engineering grows from this single idea.

Tokens and context windows: the model's desk space

The model doesn't read whole words the way you do. It chops text into small chunks called tokens. A token is roughly three-quarters of an English word, so "prompt engineering" is about three tokens, and a tidy rule of thumb is that 100 tokens equals around 75 words. The model counts everything in tokens: your question, the conversation history, and its own reply.

Now picture the model working at a desk. The context window is how big that desk is, the total amount of text (in tokens) the model can keep in front of it at once. This is its short-term memory for your conversation. Everything you have said so far, plus everything it has answered, has to fit on that desk. Modern models have huge desks, but they are not infinite.

Here's the catch. When a long chat overflows the desk, the oldest papers slide off the edge. The model literally cannot see the things that scrolled out of its context window anymore, which is why a marathon conversation can suddenly "forget" a detail you mentioned an hour ago. It isn't being careless; that text is simply off the desk. Knowing this changes how you work: for important details, restate them when they matter, or start a fresh chat with a clean summary.

ChatGPT, Claude & Gemini at a glance

The three models you'll use most in this course are OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini. They are built on the same next-word-prediction foundation, so they feel similar, but each has a personality and a sweet spot shaped by how it was trained and what it connects to.

ChatGPT is the versatile all-rounder. It is fast, creative, great at brainstorming, casual drafting, coding help, and it has a rich ecosystem of plugins, image generation, and voice. If you want one tool to reach for first, this is a safe default. Claude shines on long, nuanced writing and careful reasoning. It tends to produce thoughtful, well-structured prose, handles very large documents gracefully, and is a favorite for editing, analysis, and tasks where tone and care matter. Gemini is deeply woven into Google's world; it is strong at pulling in fresh information and lives comfortably alongside Search, Docs, and Gmail, making it handy for research and anything connected to your Google workflow.

None of these is simply "the best." The differences are real but modest, and they shift with every update. The skill that lasts is knowing how to match a model to a job, which is exactly what you'll practice next.

Choosing the right model for the task

The fastest way to pick a model is to ask one question: what does this task most need? If it needs up-to-the-minute facts or ties into your Google account, lean Gemini. If it needs long, careful, well-toned writing or analysis of a big document, lean Claude. If it needs quick creative output, coding help, or a general workhorse, ChatGPT is a dependable first stop. When stakes are high, run the same prompt through two models and compare; their disagreements are often where the interesting truth hides.

Notice how a clear ask helps any model, while a fuzzy one wastes the strengths of all three. Compare these two:

Tell me about renewable energy.

I'm writing a 300-word intro for a high school blog. Explain how solar and
wind energy work in simple, friendly language, give one real-world example of
each, and end with a single encouraging sentence about the future.

The second prompt sets the audience, length, format, and tone, so every model has a clear target to predict toward. You'll go deep on this in later lessons; for now, just feel how much the framing matters.