AI for Beginners

What AI actually is, in one minute

At the core of today's popular AI assistants — ChatGPT, Gemini, Copilot and the rest — is a model that predicts what comes next. It learned patterns from an enormous amount of data.

That single fact explains almost everything people find confusing about them.

You do not need to know more than this to use them well. But knowing this much changes how you read their answers, which is the whole difference between someone who gets value from AI and someone who gets burned by it. The slightly longer version →

What it is good at, and what it is not

Most disappointment with AI comes from handing it the wrong kind of job. The dividing line is unusually clean:

A serviceable rule: use AI to shape and explain. Check the original source for facts that matter. Which to open, task by task →

Your first ten minutes

There is no setup to learn and no command to memorise. You type a request in ordinary words, the way you would ask a capable but literal-minded assistant.

Ten minutes of that teaches more than an hour of reading about it. What a prompt is → · How to ask better questions →

Four things worth knowing before you rely on it

1. It can be wrong while sounding certain

This is the failure that catches people, because there is no tone change to warn you. A confident paragraph containing an invented date reads exactly like a confident paragraph containing a real one. The working rule that holds up: use it to understand, not to decide. Anything with money, health or law attached gets checked against an original source before you act. What it gets wrong, and what it keeps of what you type →

2. What you type is stored — and by default, used

Your conversations are kept in your account, and consumer chats may be used to help improve the models unless you turn that off. It takes four clicks. Separately, a short list of things should never be typed into any AI assistant: passwords, verification codes, bank and card numbers, and identity documents. The setting, and the list →

3. It is a tool, not an oracle — and not the only one

The most common mistake is asking AI for something a search engine does better, then concluding AI is useless. The second most common is trusting a fluent answer because it was fluent. Knowing which to open is most of the skill. ChatGPT vs Google, task by task →

4. Criminals have the same tools

This is the part of AI most likely to actually cost you something, and it has nothing to do with using a chatbot. A short voice recording can be enough to imitate someone on the phone, and AI can make convincing fake images. The defences are simple, but they work best when agreed in advance. Voice cloning, and the family code word → · Spotting AI images →

Words you will keep running into

None of these are as complicated as they sound, and knowing six of them removes most of the feeling of being locked out of the conversation.

• Prompt — what you type. That is all it means. A "good prompt" is just a clear request. More →
• Model — the trained system doing the predicting. GPT, Gemini and Claude are models; ChatGPT is the product you use one through. New versions arrive constantly, and you can safely ignore most of the announcements.
• LLM (large language model) — the kind of model behind these assistants. "Large" refers to how much text it learned from. More →
• Hallucination — the industry's word for the system stating something untrue with full confidence. An unfortunate term, because it implies a malfunction; it is closer to the ordinary behaviour of a machine built to produce plausible text.
• Training data — the text a model learned from. When people ask whether their conversations are "used for training", this is what they mean.
• Agent — an AI given the ability to take actions rather than only produce text: booking, buying, sending. This is where genuinely new risks begin, and it is worth knowing whether something you use is one.

Four things people get wrong

"It searches the internet for the answer." Not by default, and not in the way you would assume. It can look things up now and show links, but its first instinct is to generate an answer from what it learned. A citation proves it looked, not that it understood.

"If it sounds confident, it is probably right." Confidence is a writing style here, not a signal. The tone is identical whether it is on solid ground or inventing.

"There is a right way to phrase things, and I do not know it." There is no secret vocabulary and no magic words. Clear ordinary English outperforms almost every "prompt formula" being sold, and the collections of "500 perfect prompts" are largely a business model.

"Everyone else already understands this." They do not. Much of what is confidently said about AI in public is wrong, and most people using it daily could not explain what it is doing. Starting now costs you very little.

Common questions

These guides are written for adults who are new to AI and would rather not be talked down to. They are updated as the tools change; each page carries the date it was last reviewed.