AI glossary
AI terms, explained simply.
The key terms around artificial intelligence at work – short, clear and linked to the matching guide article.
A
- AI agent
-
An AI system that breaks a task into several steps and uses tools on its own to complete them – such as a web search, a file or an interface. A human should approve important steps.
Learn more : AI agent - AI literacy
-
The knowledge and skills needed to use AI systems competently and to assess their opportunities and risks. The EU AI Act addresses AI literacy in Article 4.
Learn more : AI literacy - AI policy
-
An internal rule set that defines which AI tools are allowed, which data may be entered and who reviews and is responsible for results.
Learn more : AI policy - API
-
A technical interface through which software communicates. Through an API, AI models can be integrated into your own applications and workflows.
- Audit log
-
A record that traceably captures who performed which action and when. In companies, it supports oversight, troubleshooting and evidence.
B
- Bias
-
Systematic skews in a model’s data or outputs, for example when certain groups are disadvantaged. Bias can be reduced but rarely ruled out entirely.
Learn more : Bias
C
- Chatbot
-
An application you write or speak to in natural language. Modern chatbots are usually based on a large language model.
Learn more : Chatbot - Context window
-
The amount of text a language model can consider at once in a request – prompt, history and attached documents combined. It is measured in tokens.
Learn more : Context window
D
- Deep research
-
A way of working in which AI searches many sources, consolidates the results and produces a report with citations. The sources should be spot-checked.
Learn more : Deep research - DPA (data processing agreement)
-
A contract under Art. 28 GDPR that a company concludes with service providers processing personal data on its behalf – for example, an AI provider.
Learn more : DPA (data processing agreement)
E
- Embedding
-
A series of numbers that represents the meaning of a text passage. With embeddings, a system can find relevant passages even when different words are used.
Learn more : Embedding - EU AI Act
-
The European Regulation (EU) 2024/1689 on artificial intelligence. It classifies AI systems by risk and sets out phased obligations for providers and deployers.
Learn more : EU AI Act
F
- Few-shot prompting
-
A prompting technique in which you give the model a few examples of the desired answer. This helps it match format and tone more reliably.
Learn more : Few-shot prompting - Fine-tuning
-
Further training an existing model on your own example data to adapt it to a specific task or style. For company knowledge, RAG is usually simpler.
Learn more : Fine-tuning
G
- GDPR
-
The General Data Protection Regulation (EU) 2016/679. It governs when and how personal data may be processed – including when using AI.
Learn more : GDPR - General-purpose AI model
-
A model that can be used for many different tasks, such as a large language model. The EU AI Act sets out specific obligations for providers of such models.
Learn more : General-purpose AI model - Generative AI
-
AI that creates new content – text, images, audio, code or video – instead of only sorting or scoring existing data.
Learn more : Generative AI
H
- Hallucination
-
A convincingly worded but false or invented statement from a language model, such as a non-existent source or figure.
Learn more : Hallucination - Human in the loop
-
A principle whereby a human reviews and approves AI results before they take effect – for example, before an email is sent or a contract is signed.
Learn more : Human in the loop
I
- Inference
-
Using a fully trained model to generate an answer. With cloud services, inference is often billed by the tokens consumed.
Learn more : Inference
K
- Knowledge base
-
An organized collection of company documents that an AI is allowed to access to answer questions with internal knowledge.
Learn more : Knowledge base
L
- LLM (large language model)
-
An AI model trained on very large amounts of text that can understand, summarize, translate and generate text. It calculates which words are likely to come next.
Learn more : LLM (large language model)
M
- Machine learning
-
A branch of AI in which systems learn patterns from example data instead of having every rule programmed individually.
Learn more : Machine learning - Multimodal
-
Describes models that can process several types of content, such as text, images, audio or documents within one request.
Learn more : Multimodal
P
- Personal data
-
Any information relating to an identified or identifiable person – from a name and email address to a customer number.
Learn more : Personal data - Prompt
-
The input you use to tell an AI what to do. A good prompt states the role, task, context and desired output format.
Learn more : Prompt - Prompt injection
-
An attack in which hidden instructions in text, web pages or files try to trick an AI into unwanted actions. Permission concepts and human approvals provide protection.
Learn more : Prompt injection
R
- RAG (retrieval-augmented generation)
-
A method in which an AI first retrieves relevant passages from approved documents and bases its answer on them – ideally with a citation.
Learn more : RAG (retrieval-augmented generation)
S
- Shadow AI
-
The use of AI tools without the company’s knowledge or approval, for example via private accounts. It poses risks to data protection and trade secrets.
Learn more : Shadow AI - System prompt
-
A persistent base instruction given to a model before every conversation – such as an AI assistant’s role, tone or fixed rules.
Learn more : System prompt
T
- Temperature
-
A setting that controls how varied a model’s answers are. Low values produce more consistent answers, high values more creative ones.
Learn more : Temperature - Token
-
The smallest unit of text a language model works with – often part of a word. Tokens determine how much text fits into a context window and how usage is billed.
Learn more : Token
Z
- Zero-shot prompting
-
A request without examples: the model has to solve the task from the description alone. This is often enough for standard tasks; examples help with fixed formats.
Learn more : Zero-shot prompting
Free initial consultation
Your AI start, planned together.
Show us your tasks. We’ll discuss data paths, rules and the right starting point for your team – no feature tour.
The initial consultation is free of charge. A phone number is not required.