Start Here: What Is an AI Chatbot? (And How to Think About It)

If you’ve never opened an AI chatbot before, you might feel a mix of curiosity and hesitation. By the time you finish this page, you’ll understand exactly what’s happening behind the curtain, and you’ll have tried your first real prompt. No prior knowledge needed.

The Eager Assistant Who Never Sleeps

Imagine a friend who has read a vast chunk of the internet – books, articles, code, forums – but has never experienced anything. They’ve never felt rain, been hungry, or had a job. All they can do is predict, with remarkable accuracy, what word should come next in a sentence. That friend is DeepSeek.

DeepSeek is a large language model (LLM). It doesn’t know facts the way you know your phone number. Instead, it sees patterns in language. When you type a prompt, it calculates the most probable sequence of words that would follow, based on everything it absorbed during training. It’s optimising for plausibility, not truth – which is why it sometimes sounds convincing but is factually wrong.

What makes DeepSeek special? It’s exceptionally strong at logic, math, and coding, has a huge memory (up to 1 million tokens – roughly an entire book trilogy in one go), and a reasoning mode that lets it break down complex problems. And it’s free, with no paywall. Everything you learn here works with ChatGPT, Claude, Gemini, and any modern AI assistant – but DeepSeek is the one we’ll use as our example.

The Two Mental Models That Will Save You

You don’t need to understand the math. You just need two simple ways of thinking about AI.

Model 1: The Improv Partner

If I’m an improv actor on stage, and you shout “Doctor, the patient is turning green!”, I will immediately play along as a doctor, inventing plausible medical jargon. I won’t stop to tell you I’m not a real doctor – my job is to keep the scene going. The same thing happens when you ask an AI for a biography of a historical figure: it’ll generate a fluent, plausible account. It might be correct. It might not. The improv partner model reminds you that the AI will always try to continue the pattern you started, even if that means making things up.

Model 2: Predictive Text on Steroids

Your phone’s keyboard suggests the next word as you type. An AI is that, but scaled up to entire paragraphs. If you type “The capital of France is”, the training data heavily associates that phrase with “Paris”, so it’ll likely say “Paris”. If you ask a nonsense question, it will still try to produce a reasonable‑sounding answer, because that’s what its training taught it to do.

DeepSeek’s reasoning mode adds a crucial upgrade: when activated, it thinks to itself in a structured way before answering. This doesn’t make it infallible, but it dramatically improves accuracy on logic, math, and multi‑step problems. Later, you’ll learn how to trigger this superpower with specific prompts.

Hallucination: A Feature Disguised as a Bug

When an AI confidently states something that isn’t true, that’s often called a “hallucination.” But it’s not a random error – it’s the default behaviour of a model that’s been asked to generate without constraints. If you ask, “What did Professor Jane Smith’s 2021 paper on Martian agriculture conclude?”, and I’ve never seen that paper, I won’t say “I don’t know” – I’ll invent a title, a conclusion, maybe even a citation, because that’s what the pattern of an academic answer looks like.

The solution is in your hands. You’ll learn techniques like Source Anchoring and Confidence Prompting (see the Truth & Accuracy page) that turn an AI from a confabulator into a reliable researcher. For now, remember: If accuracy matters, never accept the first answer without a verification step.

Try It Now: Your First Conversation

Open DeepSeek (or any AI chatbot) and paste this exact prompt:

Notice what just happened. You set a role (“friendly explainer”), a format (explain then confidence section), and a constraint (analogy for a 12‑year‑old). That’s the seed of the prompting framework you’ll master next.

In a Nutshell

• An AI chatbot is a pattern‑matching engine, not a database. It completes your prompts plausibly, not necessarily truthfully.
• Use the “improv partner” and “predictive text on steroids” mental models to stay appropriately skeptical.
• Hallucination is just the AI filling gaps with invention; you can drastically reduce it with verification steps.
• Your first prompt already contained the three core ingredients of a good instruction: role, format, and constraint.