An open book with a luminous human profile formed by a network of light in a library.

Getting started with AI

Published: 2026-10-10  |  DenfiS Editorial Team

Artificial intelligence is no longer just a subject for films and technology news. We can now talk to it, ask for explanations and use it to create content. But what is it really? What can it do for us — and why shouldn’t we trust it simply because it sounds convincing?

This guide opens the door to the world of AI, without complicated terminology or the need for a technical background. You’ll discover its basic capabilities and limitations, learn how to approach the answers you receive and understand why your own judgment still matters. The aim is to make the technology easier to understand, so you can recognize where it may help you and when further checks are needed.

Contents

Understand what you are interacting with

Artificial intelligence, often abbreviated to AI, is a broad term for systems that infer outputs from the data they receive, such as predictions, content, recommendations or decisions. These systems can have different degrees of autonomy. AI is therefore more than a window where you ask questions and receive text responses. [S01]

To begin, however, it helps to focus on that familiar way of working: you enter a request and the system proposes a response. The application you interact with may also be connected to search, documents or other tools. Do not assume it has access to everything you need. First check what the particular application can actually receive and process. A button label or a confident statement in a response is not enough to establish that an action has been completed successfully.

Generative AI creates content rather than simply displaying a response written in advance. A language model, the part of the system that processes and generates language, uses patterns learned during training to produce a continuation of text. This can yield a useful answer, but also convincingly incorrect content. [S02] Treat it as a proposal that still needs review: consider it, assess it and decide what to accept.

For your first attempt, choose a short text that you understand and that contains no sensitive data. Save the original and ask for a clearer version that preserves its meaning. When the response arrives, compare each claim with the original, identify additions and check whether an uncertain statement has become a definite one. If it has, request a correction to that specific passage, then read the entire text again. You can accept the result when it is easier to understand and has preserved the information, limitations and purpose. You do not need specialist commands to start: describing your goal in everyday language is enough to begin a manageable task.

Back to contents

Choose a suitable task and prepare your input

Your initial task should be small enough for you to review the entire result. Give preference to an activity whose quality you can assess and where you can reject a poor proposal without serious consequences. If you understand neither the question nor the possible answer, you will struggle to establish whether the system is helping or merely creating an impression of progress. Clarify the problem first, then use AI to work on the part you can oversee.

Separate what you want to learn from what you want to produce. Asking for an explanation and asking for finished material involve different requirements. An explanation should help you understand; material should serve a particular purpose. If you do not specify that purpose, the system may give you a response that feels complete but is of no use to you.

Before entering anything, consider which data are actually necessary. Not every detail in a document matters to the task. If you can describe the goal without names, identifying information, passwords or confidential business information, leave them out. Take particular care with other people's data: having access to them yourself does not mean you should pass them to every application.

Removing names is no reliable guarantee of anonymity. A person may remain identifiable through other details or a combination of them. [S08] Do not treat superficial changes to the text as proof that your input is safe. If you cannot assess how sensitive the content is, reduce what you submit, use a general description or seek appropriate approval before sending it.

Review your attachments too. If the task depends on a file, check that it is the right version and contains the part you want to process. Make clear which document should be used. A response based on the wrong input may be well written, but it does not solve your problem. Preparing the input is part of the task, rather than a minor technical step before the conversation.

Back to contents

Make your request clear

Start your request with the goal. State what you want to receive, who the result is for and what needs to be preserved. Context is information that helps the system understand the circumstances of the task. An instruction tells it what to do. Keeping the two clearly separate is useful, especially when you include a long text in the conversation.

Then set the boundaries. If a piece of information must not change, say so. If you want plain language, a particular length or a particular response format, state that in ordinary words. Do not add numerous conditions merely to make the request more complex. Conditions that follow from your purpose matter more than those you add because you have seen them elsewhere.

When a task involves processing supplied content, specify how that content should be treated. Should the system only reorganise the text, or may it suggest new information? Do you want a summary of an existing document or an additional explanation of the topic? Without that boundary, it becomes harder to distinguish what came from your input from what was added.

If important information is missing, ask for the gap to be identified or for a question to be asked. This is a useful instruction, but it does not remove the need for review. You still need to notice whether the response is based on the data you supplied. If a precise detail appears that you do not recognise, check where it came from before accepting it.

Your request does not have to be perfect immediately. You can use the first response to establish whether you described the goal clearly enough. If the result misses the purpose, specify what needs correcting. Explain the difference between the response you received and the one you need. You do not have to repeat the entire task from the beginning if the problem concerns one part, but restating a condition that must be preserved can help.

Back to contents

Review the response and check important claims

Begin your review by comparing the result with your request. Read it and establish whether the requested work was done, important information was preserved and anything is missing. Only then consider style. It is easy to accept fluent writing even when it changes the meaning or avoids part of the task. A good impression must not replace that initial review.

If you are processing a supplied document, compare important parts of the response with the original. Pay particular attention to passages where an omitted qualification, a changed number or a different order could affect the meaning. Do not expect a very short summary to retain every nuance. Assess whether the omitted content can also be left out for the purpose you will use the result for.

Keep fact-checking separate from language editing. Changing the wording does not establish that a claim is true. For an important statement, find an appropriate external source, check that it addresses the specific issue and pay attention to its date and limitations. If the system supplies a reference, open it and establish whether it actually supports what was written. Checking sources and citations forms part of the recommended oversight of AI outputs. [S02]

Asking the same system another question may help you obtain clarification, but do not count it as independent confirmation. If a response affects an important decision, you need a check that does not depend solely on the system repeating itself. When you cannot verify an important claim, the most honest conclusion is that it remains unconfirmed. There is no need to fill uncertainty with guesses.

Back to contents

Decide where to draw the line

Use an AI proposal only to the extent that you can take responsibility for applying it. With text, this means understanding what it says and accepting its meaning. With a plan, it means that circumstances and constraints have been considered. With a task that carries serious consequences, it means involving a qualified professional rather than replacing their judgement with an application conversation.

Pay particular attention to the difference between a proposal you receive and an action that has been performed. You may have asked only for material to be prepared, while the application also offers a connection to another system. That creates a new decision about access and action. Before continuing, establish what you are allowing, which data it concerns and what the consequences could be. Getting started does not require approving every access option offered to you.

A successful first step is not measured by how much the response surprises you. It is measured by whether you now understand the task more clearly, have a usable result and know its limits. You can continue on your own when you can describe the goal, prepare suitable input, review the response and recognise what needs external verification. That foundation remains useful even when you change applications.

Back to contents

If you’d like to learn more about this topic, download the free ebook “Everything you’ve always wanted to know about artificial intelligence but didn’t know whom to ask...”. Discover how AI works, how it can help you in everyday life, learning and work, and what to watch out for when using it.

Selection Methodology

This guide is based on the first chapter of the book “Everything you’ve always wanted to know about artificial intelligence but didn’t know whom to ask...”, by Dejan Popović, DenfiS. The chapter was prepared as a standalone guide for beginners. Its text and the order of its five thematic sections have been preserved here.

The suggested approach takes you from a small task you can assess to reviewing the result and deciding how to use it. The practical recommendations are an editorial way of organising your work, rather than a promise that the same approach will produce reliable results in every application or situation. This guide does not rank products or declare one tool the best for every task.

The labels S01, S02 and S08 refer to the sources listed below. These references come from the book’s bibliography. The official sources were checked again on 10 October 2026 for the claims cited in this guide.

Related Categories

Related AI Tools

Frequently Asked Questions (FAQ)

Do I need to know how to code to get started?

For the initial task described in this guide, you can explain your goal in everyday language, provide suitable input and review the result. You do not need specialist commands.

What should I choose for my first attempt?

Choose a short text that you understand and that contains no sensitive data. Save the original, ask for a clearer version that preserves the meaning, then compare the response with the original.

What should I include in my request?

Explain what you want to receive, who the result is for, what must be preserved and what response format suits you. If you provide a text, specify whether additions are allowed or whether you want only the existing content to be processed.

Does removing names make a document anonymous?

Do not assume so. Other details, or a combination of them, may make a person identifiable. If you cannot assess how sensitive the content is, use a general description or seek appropriate approval before sending it. [S08]

Does a well-written answer mean it is correct?

No. First check whether it meets your request and preserves the information in your input. Verify important factual claims using appropriate external sources. Fluent writing is not proof of accuracy. [S02]

What should I do if the answer is not useful?

Explain specifically what does not suit your purpose and what needs correcting. Restate the condition that must be preserved. After the correction, read the entire result.

Is asking the same AI again enough to verify an answer?

It may help clarify the response, but it is not independent confirmation. Check an important claim against an appropriate source. If you cannot verify it, keep it marked as unconfirmed.

Do I have to allow access to my email or other systems?

The initial task in this guide does not require that access. If an application offers a connection, assess separately what you are allowing, which data it can access and which actions it can perform.

When should I involve a professional?

When the result could have serious consequences or you do not have enough knowledge to assess whether it is correct. AI can help prepare questions or materials; involve a professional before applying the result in an important situation.

This guide is educational material. For tasks with serious consequences, involve an appropriate professional. Check the capabilities of the particular application before use; its response alone is not proof that an action has been performed.

All Guides