AI terms, explained simply
The most important AI terms — clear, jargon-free and hype-free. Made for decision-makers in mid-sized companies, not for computer scientists.
- Artificial Intelligence (AI)
- Software that handles tasks that used to require human thinking — understanding language, writing text, spotting patterns in data. AI doesn't "think" like a human; it works with probabilities.
- Machine Learning (ML)
- A subfield of AI: instead of fixed rules, the system learns patterns from example data. The more good data, the better the results.
- Large Language Model (LLM)
- The "language model" behind tools like ChatGPT. Trained on huge amounts of text, it predicts the most likely next word — remarkably useful for text, summaries and answers.
- Generative AI
- AI that creates something new: text, images, code or speech. The umbrella term for the current AI wave.
- Prompt
- The input or instruction you give the AI. The clearer the prompt, the better the result — but perfect "prompt engineering" is usually overrated.
- Token
- The smallest processing unit of a language model — roughly a word fragment. AI usage is often billed per token, so tokens drive running costs.
- Hallucination
- When AI states something convincing but factually wrong. The main reason important AI output always needs a human check.
- RAG (Retrieval-Augmented Generation)
- A method where the AI draws its answers from your own documents instead of just its training knowledge. Answers become precise, current and verifiable. More on this: Application examples →
- Embeddings & vector database
- The tech behind RAG: text is translated into numbers ("embeddings") and stored in a vector database so the AI can find relevant passages instantly.
- AI agent
- An AI that doesn't just answer but acts: plans several steps, looks things up in systems and completes tasks — ideally with human approval at critical points. More on this: When agents pay off →
- Chatbot vs. AI agent
- A chatbot answers questions. An agent also performs actions (e.g. fetch data, enter, route). The difference decides the real value.
- Fine-tuning
- Re-training a model on your specific data or style. Often unnecessary — RAG is frequently enough and cheaper.
- Process automation
- Letting software handle recurring workflows. With AI you can automate even tasks that used to need human judgement. More on this: 6 examples with ROI →
- API
- An interface that lets programs talk to each other. It's how you embed AI into existing tools (CRM, ERP, website) instead of building an island.
- EU hosting & on-premise
- Where your data is processed. EU hosting (servers in the EU) or on-premise (on your own systems) is the key to GDPR-ready AI.
- GDPR & AI
- AI isn't automatically a data-protection problem. With EU hosting, business plans and walled-off solutions, AI can run compliantly. More on this: Use ChatGPT safely →
- Prototype / MVP
- A first working minimal version to test idea and value fast — before investing big. The most pragmatic entry into any AI project.
- Open-source vs. proprietary models
- Open models can be self-hosted (more control, privacy); proprietary ones (e.g. from OpenAI) are often more capable but run externally. The choice depends on the use-case.