Artificial intelligence (AI) is the field of computing that builds programs able to do tasks that used to need human intelligence: recognizing a face, translating, recommending a song or holding a conversation. Instead of following only hand written rules, most of them learn patterns from many examples.
A question from the app. Pick an answer.
If it answers confidently, is it true?
It can make things up in the same confident tone. That is called a hallucination.
Its tone does not show whether it knows: it can make things up just as calmly. Sounding sure is not being right; check what matters.
The details can be made up too. Being specific does not make it true.
AI is not just ChatGPT. It is the filter that sends an email to spam, the bank that blocks an odd purchase on your card, the map that estimates your travel time in traffic and the phone keyboard that guesses your next word. Many of these AIs do not generate anything: they classify, predict or recommend.
The kind that chats and writes text is one specific type: generative AI. It is the most visible today, but it is only one part of a much larger field.
Say you want to catch spam. With hand written rules you would write: "if the email says 'prize' or 'free', it is spam". That works for a while, until spammers write "pr1ze". With machine learning you give the program, for example, 10,000 emails already labeled as spam or not spam, and it finds the patterns on its own.
That process is called training, and its result is a model. How good the model is depends a lot on the examples: if it barely saw cases of something, that is where it tends to fail.
A simple way to sort them is by what they do. They classify: spam or not spam. They predict: how long your trip takes or whether a purchase is fraud. They recommend: the next show or the next video. And they generate: text, images, code or voice, like language models (LLMs), built with neural networks.
Many are good at one task; language models do several. There is no agreement on whether that already counts as "general" AI, the kind that learns anything the way a person does.
Many models were trained on more English text than other languages, and on fewer examples of each regional variety. So they can misread a local expression or explain a procedure using another country's rules: you ask how to renew your ID and get the process for a different country. If the question is local, say which country you mean and check an official source.
"AI thinks like a person." It imitates patterns very well, but sounding human does not prove it understands the way you do. "AI is a robot." Almost always it is software with no body. "If an app says it uses AI, it is better." Not necessarily: sometimes it is a trendy label for ordinary rules.
To understand the AI that chats, go in this order: what an LLM is, what a token is, what a prompt is and why it sometimes makes things up. Each term comes with a question to practice.