AI glossary
AI glossary: large language model, prompt, hallucination, AI agent, RAG, the AI Act and more, explained in plain language.
A
- AI Act
- The EU regulation on artificial intelligence that classifies AI systems by risk level and sets obligations for providers and users. Its obligations apply in stages.
- AI agent
- An AI system that carries out several steps on its own to reach a goal, for example finding data, using tools and preparing a result. An agent works within the limits you set.
- AI assistant
- An AI tool that helps with everyday work, such as drafting emails, summarising documents or finding information. Unlike an agent, it usually waits for your instruction at each step.
- AI literacy
- A basic understanding of what AI can do, where its limits are and how to use it safely. The AI Act expects companies to ensure adequate AI literacy among their staff.
- AI process automation
- Using AI to carry out repetitive steps in a workflow, such as sorting emails or preparing reports. The goal is to save time and reduce errors.
- AI strategy
- A plan that defines where and how a company will use AI, with which goals, resources and rules. It connects business priorities with concrete steps.
- Artificial intelligence (AI)
- Computer systems that perform tasks that usually need human thinking, such as understanding text, recognising patterns or making decisions. In companies, AI most often helps with writing, analysis and automating routine work.
B
- Bias
- A systematic error where AI gives unbalanced or unfair results because of the data it learned from. For important decisions, a person should always review the results.
C
- Context window
- The amount of text an AI model can take into account at once, including your prompt and attached documents. If there is too much content, the model may overlook parts of it.
F
- Fine-tuning
- Additional training of an existing AI model on specific data so it performs a particular task better. A good prompt or RAG is often enough, so fine-tuning is not always needed.
G
- GDPR and AI
- Data protection rules that also apply when using AI tools. Do not enter personal or confidential data into tools without proper contractual and technical safeguards.
- Generative AI
- A type of AI that creates new content: text, images, code or summaries. It works from patterns learned from large amounts of data.
H
- Hallucination
- An AI answer that sounds convincing but is wrong or made up. That is why important facts from AI tools should always be checked.
- Human-in-the-loop
- The principle that a person reviews or approves an AI output before an important step. This reduces the risk of errors and keeps accountability clear.
K
- Knowledge base
- An organised collection of company documents, guidelines and data that an AI tool can access. A good knowledge base is a precondition for reliable answers from an internal AI assistant.
L
- Large language model (LLM)
- An AI model trained on huge amounts of text that can understand and produce language. Tools such as ChatGPT, Claude or Gemini are built on such models.
M
- Machine learning
- A field of AI where systems learn from data instead of hand-written rules. It is used, for example, to forecast demand or detect anomalies.
- Multimodal model
- An AI model that understands and produces several types of content, such as text, images, audio or video. Such a model can, for example, describe a photo or read a table from an image.
P
- Pilot project
- A small, time-boxed trial of an AI solution on a selected process or team. It shows whether the solution works before you commit to a wider rollout.
- Prompt
- The instruction or question you type into an AI tool. The clearer and more specific the prompt, the more useful the answer.
- Prompting
- The skill of writing instructions for AI tools so you get accurate, useful results. It includes giving a role, context, an example and the desired output format.
R
- RAG (Retrieval-Augmented Generation)
- An approach where the AI model first looks up information in your documents or knowledge base before answering. This makes answers more accurate and grounded in your own data.
- Responsible AI
- Using AI in a way that is safe, fair, transparent and compliant. It includes human oversight, checking outputs and clear internal rules.
T
- Token
- The basic unit of text a language model processes, roughly part of a word. The number of tokens affects usage cost and how much text the model can handle.
U
- Use case
- A specific task or process where AI brings a measurable benefit. Choosing the right use cases is the first step to a successful AI rollout.