Learning

    Questions about AI

    You do not need to know the technology to start understanding AI. Find straightforward answers about AI at work, without the hype.

    This is just a taste. Zaipiens Community offers more knowledge, practical tips and tools – and above all, connections with fellow Zaipiens.

    Become a Zaipiens – discover the Community →

    26 questions shown

    Basics

    What is an LLM?

    A large language model learns language patterns from large amounts of training data. It builds text step by step using context. For example, it can draft a meeting agenda. A convincing answer is not proof that its content is true.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    How is an AI service different from a language model?

    An AI service has an interface and features built around one or more models. The language model generates text; the service may also provide file or web search. Available features depend on the service, plan and settings. The same model can power several different services.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    What does generative AI mean?

    Generative AI creates content such as text, images or audio using patterns in training data and your instruction. Language models are part of this field. For example, ask for three titles for an imaginary course. You still need to judge whether the suggestions fit.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    What is a token?

    A token is a small piece of text, sometimes a whole word and sometimes part of one. Models process text as tokens. Counts vary by language and text, so one word does not always equal one token. Limits and API prices often use tokens.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question

    Practical use

    What is a good first work task?

    Choose something small that you can check yourself. Ask for an agenda for an imaginary project meeting or a clearer version of public text. Compare the result with your goal and edit it. Do not start with sensitive customer information or decisions affecting people.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    How can I use AI to summarise?

    Use only material allowed in your organisation's approved tool. Ask for five main points and questions the text leaves unanswered. Then check the original for important conditions and exceptions. A summary can omit or invent information.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    How do I know whether AI helps?

    Compare a few similar tasks with and without AI. Include time spent giving instructions, checking and correcting, not just waiting for an answer. Assess quality and risk too. For example, is a draft public instruction clearer after your review?

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question

    Prompting

    How do I write a clear instruction using PKFI?

    In Zaipiens training, we use PKFI to help write and improve AI instructions. We keep the Swedish acronym: Persona, Kontext (Context), Format and Iteration. P – Persona: What role or perspective should the AI take? For example, a meeting facilitator. A role does not make the AI a real expert. K – Context: Describe the task, goal, audience and relevant background. Only use material you are allowed to share in the tool. F – Format: Specify how the answer should look, such as three points or a table. I – Iteration: Read the answer, check it and give feedback. Improve the instruction step by step. Example of a first instruction: “Act as a meeting facilitator. Help an imaginary project team plan a 30-minute meeting to choose next steps. Write an agenda with three items and timings. Ask if essential information is missing.” After reading the answer, you can iterate: “Allow more time for discussion, but keep the meeting within 30 minutes.” Check that the timings add up. PKFI is a guide, not a guarantee of correct answers.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    Does giving examples help?

    An example can show tone, length and structure more clearly than a general request. Show a short public heading and ask for three similar options on another topic. Check that the model has not copied inappropriate details from the example. Do not use confidential examples.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    What if the answer is poor?

    Use the I in PKFI: Iteration. Explain what is missing and change one thing at a time. Do you need to clarify the Persona, Context or Format? For example: “Shorten this to three points and explain the jargon for a beginner.” Read and check the new answer. If it contains errors, return to the original source. Asking the model to try again does not guarantee greater accuracy.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question

    Tools and concepts

    What are Skills in an AI tool?

    Skills are reusable instructions and resources for a particular task. A skill might describe how the AI should summarise a meeting using your template. In the open Agent Skills format, instructions live in a file named SKILL.md, sometimes alongside templates or code. This is not the same as retraining the model. Support varies between tools. Read the contents and check the source before using a skill, especially if it can run code.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    What is a project in an AI tool?

    A project can bring together work on a topic or task. In Claude, for example, projects can contain instructions, files and conversations that give the AI context. You might collect public course material and instructions about the audience in a training project. How information is reused, stored and shared depends on the service and settings. A project does not automatically mean the AI remembers everything or that all material is private. Check access and what you are allowed to upload.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    What is an artifact?

    An artifact is an output you can work on separately from the conversation, such as a document, a diagram or a small interactive tool. Claude calls this feature Artifacts. One example is a checklist you ask the AI to create and then revise. Supported types and sharing options vary between services. An artifact is not automatically fact-checked or safe to use. Check its contents, any code and its sharing settings before using or distributing it.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    What is a plugin?

    A plugin is an add-on that extends a program's features. In AI tools, it may bundle skills and connections to other services. Claude Cowork, for example, can package these components together as a plugin. Providers use the term differently; a plugin is not the same as a model. Check who made the add-on, what information it can read and what actions it can perform. Only install add-ons approved for your work.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    What is a .md file, and why is it used with AI?

    A .md file is a plain-text file written in Markdown, a simple way to mark headings, lists and other content. For example, “# Meeting agenda” marks a heading and “- Next steps” marks a list item. You can open the file in a text editor. It is useful for readable instructions, notes and templates, such as a PKFI template. The file extension does not make the contents true or an active AI instruction: the tool must read and use the file for that purpose. Do not put passwords or other secrets in shared files.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    What is MCP?

    MCP stands for Model Context Protocol. It is an open standard for connecting AI applications to external data sources and tools. An approved connection might let an AI app search a document source or use a calendar tool. MCP is not an AI model and does not itself grant permission to read everything. Available access and actions depend on the connection and permissions. Check the provider and limit access; actions that send or change something need clear approvals.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    What is an API, and how is it different from MCP?

    API stands for Application Programming Interface. It is an interface that lets software use another program's features or data. An app might fetch weather information or send a question to an AI service through an API. MCP provides a common way for AI applications to connect to tools; a tool may use an API behind the scenes. The terms are not interchangeable. API use may have separate costs and terms. If an API key is used, treat it as a secret, not something to paste into a chat or shared file.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question

    Reliability

    Why does AI make things up?

    A language model generates plausible text, not a guaranteed factual report. With limited evidence it may still sound certain, for example inventing a source. This is often called hallucination. Check important claims against original sources and accept that sometimes the answer should be “I don't know”.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    How do I check an AI answer?

    Identify the claims that matter. Open the cited original sources and check content, dates and context. Recalculate important numbers separately. Asking the same model whether it is sure is not an independent check. Without evidence, do not present a claim as fact.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question
    Does the model know today's news?

    Not necessarily. Training data has a cutoff, and a service may or may not offer web search. Even with search, check the source date and what it actually says. This question box does not search the web; it only uses currently reviewed FAQ material.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    Can AI give biased answers?

    Yes. Training data, instructions and the way a service is built can affect answers. Ask for different perspectives, but check what is missing yourself. Do not use an AI judgement as the sole basis for recruitment or other decisions about people.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 17 Feb 2027

    Review schedule: quarterly

    Sources to check

    Link to this question

    Privacy

    Can I paste in work material?

    Only if your workplace rules and agreements allow it in that particular tool. First check confidentiality, personal data, access rights and provider handling. Removing names is not always enough: other details can identify someone. Use fictional or public material if unsure.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    Are my questions used to train AI?

    It depends on the service, account, settings and agreement. Consumer service terms are not automatically the same as API or business terms. Check whether questions, files and answers may be used for training, how long they are kept and what choices you have. Data not being used for training does not mean it is never stored. Always check the current terms for your particular tool.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    What happens to the question I ask here?

    When enabled, your question and reviewed FAQ material are sent through Replit to OpenAI. The site does not save questions or answers in logs, CRM or history. Short-lived pseudonymous limit keys and usage counters help prevent abuse. Replit and OpenAI process information under their terms; this is not a promise of zero provider retention. Do not send personal information or secrets.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    Who is responsible for the result?

    AI does not take over your responsibility or your employer's. Decide who reviews results, what information may be used and when a person must make the decision. For example, have a responsible person check a draft before sending it to a customer. Legal or medical issues require appropriate expertise.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question

    Rules and responsibility

    What does the EU AI Act mean for people using AI?

    The EU AI Act regulates AI based on factors including its use and risk. It does not only concern developers: organisations using AI may also have obligations. Some uses are prohibited. Others may fall under high-risk requirements or duties to disclose AI use. AI used in recruitment is one example that needs particular assessment. Article 4 requires providers and organisations deploying AI to take measures supporting the AI literacy of people using systems on their behalf, taking account of the task and their existing knowledge. Attending a course does not by itself mean an organisation complies with the entire regulation. Start by mapping which AI tools you use, for what purposes and who is responsible. Decide what material may be used and when a person must review the result. Rules have different application dates and exceptions; check current EU guidance for your particular use. GDPR also applies when personal data is processed. This is an introduction, not legal advice.

    Reviewed – within the review period

    Last reviewed: 19 Nov 2026 · Reviewer: Jacob Bergström

    Next review: 19 Dec 2026

    Review schedule: monthly

    Sources to check

    Link to this question
    Zaipiens Community

    You do not have to explore AI alone.

    An answer can help you move forward. But learning with others lets you ask follow-up questions, share experiences and discover new ways to use AI in everyday work. Our Community gives you much more to learn from and try – alongside people who want to keep growing too.

    Become a Zaipiens – discover the Community →

    Go directly to the source

    Official guidance when you want to check a fact or read more.