Six lessons from zero to using AI

What AI actually is (and isn't)

Artificial intelligence is software that performs tasks we used to think required human intelligence: understanding language, recognizing images, writing text, making predictions. When people say "AI" today, they almost always mean generative AI — tools that create new text, images, audio or video on request.

Here's the key mental shift: modern AI is not programmed with rules, it is trained on examples. Nobody wrote a rule that says "after 'once upon a' comes 'time'." Instead, the system read enormous amounts of text and learned the patterns of language itself — grammar, facts, styles, reasoning steps — by predicting what comes next, billions of times over.

What AI is not: it is not a conscious mind, it is not a search engine (although many tools now search the web as an extra step), and it is not always right. Think of it as an extremely well-read, endlessly patient assistant that works fast, never gets bored — and occasionally states a wrong thing with full confidence. That last part is why Lesson 5 exists.

• Machine learning — the broad technique of letting software learn from data instead of fixed rules.
• Neural networks — the brain-inspired math structures doing the learning.
• Large language models (LLMs) — neural networks trained on text. ChatGPT, Claude and Gemini are all built on LLMs.

How ChatGPT, Claude & Gemini actually work

Every major chatbot works the same way underneath. The model reads your message broken into tokens — small chunks of text, roughly three-quarters of a word each — and then predicts the most likely next token, again and again, until it has written a full reply. That's it. The magic is that predicting language well, at massive scale, turns out to require genuinely understanding a lot about the world.

Three ideas explain most of what you'll experience:

• Training vs. chatting. The model learned during training, which ended at some point in the past (its knowledge cutoff). It doesn't learn from your chats by default. If you need fresh information, use a tool with web search switched on.
• The context window. The model can only "see" what's in the current conversation, up to a limit. Long chats can push early details out of view — if it seems to forget, start a fresh chat and restate what matters.
• It predicts, it doesn't look up. Answers are generated, not retrieved from a database. Usually that's fine; sometimes it produces a confident-sounding error called a hallucination.

Modern models are also multimodal: you can show them photos, PDFs and spreadsheets, and many can produce images or charts back. And the newest "reasoning" models will visibly think step-by-step before answering hard questions.

Pick a tool and set it up

Good news: you can't really pick wrong. The big three chatbots are all excellent, all free to start, and all work in your browser or as a phone app:

• ChatGPT (chatgpt.com, by OpenAI) — the most widely used all-rounder.
• Claude (claude.ai, by Anthropic) — known for strong writing, careful reasoning and handling long documents.
• Gemini (gemini.google.com, by Google) — deeply connected to Gmail, Docs and the rest of Google.

Pick one, create a free account, and stay with it for two weeks before trying another — fluency with one tool beats shallow contact with three. (When you're ready to compare, our model guide breaks down which is best at what.)

Two settings worth checking on day one: whether web search is available (great for current events), and the privacy/data controls, where you can usually opt out of your chats being used for training. Rule of thumb regardless of settings: don't paste in passwords, or other people's private data.

Free plans have usage limits and may use smaller models. Paid plans (around $20/month across the board) get you the strongest models and higher limits — worth it once AI becomes a daily habit, unnecessary on day one.

How to talk to an AI (prompting, minus the mystique)

A prompt is just whatever you type. "Prompt engineering" sounds technical, but 90% of it is one idea: talk to it like a capable new colleague. A new colleague needs context, a clear ask, and an example of what good looks like. So does AI.

The four ingredients of a strong prompt:

• Context. Who you are, what you're working on, who it's for. "I run a small bakery and I'm writing to loyal customers…"
• Task. The specific thing you want. "…draft a friendly email announcing our new opening hours…"
• Format. Shape and length. "…under 120 words, warm tone, with a one-line subject."
• Example or constraints (optional but powerful). Paste a previous email you liked, or say what to avoid.

Then — and this is the part beginners miss — treat the first answer as a first draft. Reply with "make it shorter," "more formal," "give me three options," "you lost the warmth, bring it back." Conversation is the real skill. You can also ask the model to interview you: "Ask me five questions before you start."

What AI gets wrong — and how to stay safe

Trusting AI blindly is the #1 beginner mistake. Here's the honest list of failure modes:

• Hallucinations. The model can invent facts, quotes, citations, even court cases — fluently and confidently. Frequency has dropped a lot in recent models, but it is not zero.
• Stale knowledge. Without web search, the model knows nothing after its training cutoff.
• Bias. Models learn from human text, and human text contains human biases. Outputs can reflect them subtly.
• Confident tone ≠ correct content. The writing quality is constant whether the answer is right or wrong, which disables your usual "this sounds off" radar.

Your protection is one habit: match your verification to the stakes. Brainstorming dinner ideas? No checking needed. A fact going into a report? Ask for sources and click them. Anything medical, legal or financial? Treat AI as a way to get oriented and prepare better questions for a professional — not as the professional.

Privacy in one line: assume anything you paste could be stored, so keep out passwords, and confidential information about other people or your employer (unless you're on a business plan that contractually protects it).

Make AI part of your week

Knowledge fades; habits stick. The goal of this final lesson is to wire AI into moments you already have. The trigger to train into yourself: whenever you think "this is going to be tedious" — drafting, summarizing, comparing, planning, translating, debugging — that's your cue to open a chat.

Where beginners reliably get value in week one:

• Writing: emails, messages, reviews, cover letters — drafted in your voice, in seconds.
• Understanding: paste a contract, policy, or dense article and ask for a plain-language summary plus "what should I watch out for?"
• Planning: trips, meals for the week, a workout plan, a project broken into steps.
• Learning: a personal tutor for any subject that never tires of "explain that more simply."
• Deciding: "Here's my situation and my options — argue both sides, then tell me what questions I should be asking."

From here, your path on this site: do one hands-on tutorial, skim the model guide so you know your options, bookmark the news page, and keep the glossary handy for any term that trips you up.