AI GUIDEUnderstand the AI system you are using
A letter is hard to follow. A family story needs putting into words. You have a rough idea for a picture but cannot draw it. These are ordinary starting points for using AI: bring something you have, describe what would help, and inspect what comes back. You do not need to learn programming first.
This opening lesson gives you a map of what is happening behind the conversation. The aim is to recognize possibilities, not memorize a new vocabulary.
AI is broader than a chat window
Artificial intelligence includes systems that recognize patterns, make predictions or generate material. A spam filter sorts messages; a prediction system estimates something; a generative assistant creates text or images. Machine learning develops models from data rather than specifying every decision as an explicit rule. The IBM overview and Google introduction explain these distinctions.
You have probably encountered some of these systems without calling them AI. Conversational applications make more tasks accessible through ordinary language. That does not make a conversation the best solution to everything: adding two numbers still needs a correct calculation, not a persuasive paragraph.
A large language model generates language using units called tokens, which can be words or parts of words. You do not need the mathematics to use one, but it helps to know that an answer is generated in context rather than selected from a guaranteed list of approved facts. See Google's language-model introduction for optional depth.
Four parts of a useful mental model
Consider: “Explain this letter in plain language and draft a reply.” The model processes the request and generates text. The application is the service you open, with its particular features and settings. The source is the letter you actually provide. A separate action would send the reply.
You can ask for a draft without allowing that action. And the assistant cannot explain your actual letter if it has never received it. Different applications can use related models while offering different file, voice or connected-service features. This is why choosing an app involves more than recognizing a model name.
Match the input to the task
Some tasks start with text; others start with a photo, recording or table. These forms are called modalities. A multimodal system can work with more than one. For example, visual question answering is documented in Google's image guide.
A photo of an appliance label might help identify the printed model number, provided it is readable. A voice conversation can support language practice. A small table can become a chart if the application has calculation and chart tools. The result you want determines the material and tools needed; a description of a chart is not the chart itself.
Walk through a small task
Use this fictional community-library notice: “The reading room will be closed on Saturday. The return box remains available. Sunday opening hours are unchanged.” Ask: “Explain this in two short sentences. Separate what is stated from anything not specified.”
An illustrative editor-written answer is: “You cannot use the reading room on Saturday, but you can still return books through the return box. The notice says Sunday hours stay the same; it does not give those hours.” The benefit is a clearer explanation without losing the exception. The limitation: the assistant cannot recover missing hours from this notice.
Now ask for a polite reply requesting the Sunday hours. You should receive a draft, not a claim that it has been sent. If the assistant invents “Sunday, 10–4,” compare that with the source and remove it. No real letter or personal record is needed for this first experiment.
Exercise
Choose the material and check for each task:
1. Rewrite an invented birthday greeting more warmly.
2. Explain the exact text of a notice you received.
3. Send your reply to its sender.
4. Assess pronunciation from a typed sentence alone.
The first needs a text draft and your judgment of tone. The second needs the actual relevant text. The third additionally needs a connected sending tool and your approval of recipient and message. The fourth lacks the audio needed for pronunciation feedback. Asking for more suitable evidence is a useful result, not a failed conversation.
Use this in work and daily life
The same distinctions scale up. A teacher can draft an exercise, a traveler can organize preferences, and a CTO can discuss a software change. None of these automatically establishes a checked answer key, a confirmed booking or a deployed change.
Try describing one task of your own: what you already have, what you want to receive, and how you would recognise a useful result. In the Base programme, continue with B01. If you chose the shorter Start from scratch route, continue to A02. The programme navigation below follows your selected route.
Sources
Read the original sources to explore the topic and check specific claims.