Models and inputs
- AI
- An umbrella term for systems performing tasks related to perception, language, pattern recognition or decision support.
- Machine learning
- Methods in which a system learns patterns from data rather than only executing fixed rules.
- LLM
- Large language model: a model using probabilities over text units to process and generate language.
- Prompt
- The input or working instruction given to an AI system.
- Token
- A small text unit used by a language model; it is not always the same as a word.
- Context window
- The amount of material a model can consider in one interaction.
Data and extensions
- Embedding
- A numerical representation that captures similarity between text, images or other data.
- RAG
- Retrieval-augmented generation: relevant material is retrieved from a knowledge source before the model answers.
- Fine-tuning
- Additional adaptation of a model with selected examples or data.
- Multimodal
- A system can process several media types such as text, image, audio or video.
- Agent
- A system that plans several steps, uses tools and processes results further.
Errors and responsibility
- Hallucination
- A confidently worded but incorrect or invented output.
- Bias
- A systematic distortion in data, model or use.
- Deepfake
- Synthetic or altered media that can simulate real people or events.
- AI literacy
- The ability to use AI purposefully, critically, safely and transparently.
Current model and tool names are available in the KAI topic hub.