KI-Glossar — Alle Begriffe erklärt

A

Agentic AI

Agentic AI refers to AI systems that act proactively and autonomously, planning and executing tasks without waiting for instruction at each step.

Fondamental

AGI (Artificial General Intelligence)

Artificial General Intelligence

AGI refers to a hypothetical AI with cognitive capabilities equal to or exceeding humans across all intellectual domains.

Technique

AI Agent

An AI agent is an autonomous system that uses an LLM to plan, decide and execute real tasks without human intervention at each step.

Éthique

AI Alignment

AI alignment aims to ensure an artificial intelligence system acts in accordance with human values and intentions.

Infrastructure

AI API

An AI API allows developers to integrate artificial intelligence capabilities into their applications.

Évaluation

AI Benchmark

An AI benchmark is a standardized test that measures and compares AI model performance on specific tasks.

Application

AI Coding

AI Coding encompasses tools and techniques that use AI to write, debug, review and deploy software code.

Comportement

AI Hallucination

An AI hallucination is a response generated by an AI model that appears plausible but is factually incorrect or fabricated.

Technique

AI Inference

Inference is the process of using a trained AI model to generate predictions or responses from new data.

Infrastructure

AI Orchestration

AI orchestration coordinates multiple models, agents and tools in a unified workflow to accomplish complex tasks.

Comportement

AI Reasoning

AI reasoning refers to a model's ability to break down a problem into logical steps to reach a conclusion, rather than answering instinctively.

Éthique

AI Safety

AI Safety is the field focused on ensuring AI systems are safe, reliable and don't cause unintended harm.

Technique

AI Voice Cloning

AI voice cloning is a technique that faithfully reproduces a person's voice from a short audio sample, enabling synthetic speech generation with the same timbre, intonation and accent.

Éthique

AI Watermarking

AI watermarking embeds invisible markers in AI-generated content to identify its origin.

Fondamental

Artificial Intelligence (AI)

Artificial Intelligence

Artificial intelligence refers to computer systems capable of performing tasks that typically require human intelligence: language understanding, reasoning, learning and decision-making.

Modèle

Attention Mechanism

The attention mechanism allows a model to weigh the importance of each word relative to all others, capturing global context.

Technique

Autonomous AI Agent

An autonomous AI agent executes complex tasks end-to-end without human intervention, planning and adapting its actions.

C

Technique

Chain of Thought

Chain of Thought is a prompting technique that asks the model to show its step-by-step reasoning before giving its final answer.

Technique

Context Window

The context window is the maximum amount of text an LLM can process in a single request.

D

Éthique

Data Poisoning

Data poisoning is an attack that injects malicious data into an AI model's training set to corrupt its behavior or predictions.

Fondamental

Deep Learning

Deep Learning is a subset of Machine Learning using multi-layered neural networks to learn complex representations from raw data.

Éthique

Deepfake

A deepfake is synthetic media content (video, audio, image) generated by AI that realistically mimics a real person's appearance or voice.

Modèle

Diffusion Model

A diffusion model is an AI architecture that generates images starting from random noise and progressively refining it.

E

Technique

Embedding

An embedding is a numerical representation (vector) of text or data, capturing its semantic meaning.

F

Technique

Few-shot

Few-shot Learning

Few-shot involves providing a few examples in the prompt to guide the model toward the desired response format or style.

Technique

Fine-tuning

Fine-tuning is the process of retraining an existing AI model on a specific dataset to adapt it to a particular domain or task.

Modèle

Foundation Model

A foundation model is a large AI model pre-trained on massive data, adaptable to multiple tasks.

Technique

Function Calling

Function Calling enables an LLM to call external functions or APIs in a structured way to execute concrete actions.

G

Modèle

GAN (Generative Adversarial Network)

Generative Adversarial Network

A GAN is a deep learning architecture composed of two competing neural networks — a generator and a discriminator — to produce realistic synthetic data.

Fondamental

Generative AI

Generative AI refers to artificial intelligence systems capable of creating original content: text, images, video, audio, code.

Infrastructure

GPU Cloud

GPU Cloud provides on-demand graphics processors for training and running AI models without hardware investment.

Technique

Grounding

Grounding connects LLM responses to factual data sources to reduce hallucinations and increase reliability.

K

Infrastructure

Knowledge Graph

A Knowledge Graph organizes knowledge as a network of relationships between entities, enabling AI to reason over structured facts.

L

Fondamental

Latent Space

Latent space is a compressed mathematical representation where an AI model encodes the essential features of data, enabling content generation and manipulation.

Modèle

LLM (Large Language Model)

Large Language Model

An LLM is an AI model trained on billions of texts, capable of understanding and generating human language.

Technique

LoRA

Low-Rank Adaptation

LoRA is an efficient fine-tuning technique that adapts an AI model by only adjusting a fraction of its parameters, drastically reducing cost.

M

Fondamental

Machine Learning

Machine Learning is a branch of AI where systems learn from data to improve their performance without being explicitly programmed for each task.

Infrastructure

MCP (Model Context Protocol)

Model Context Protocol

MCP is an open protocol that standardizes the connection between LLMs and external tools/data.

Modèle

Mixture of Experts (MoE)

Mixture of Experts

MoE is a model architecture that activates only a fraction of its parameters for each request, making large models more efficient.

Comportement

Model Collapse

Model collapse is a phenomenon where an AI model trained on data generated by other AI models progressively loses quality and diversity, converging toward degenerate outputs.

Technique

Model Distillation

Distillation transfers knowledge from a large model (teacher) to a smaller model (student), preserving performance at lower cost.

Technique

Multi-agent System

A multi-agent system coordinates multiple specialized AI agents that collaborate to solve complex problems.

Modèle

Multimodal

A multimodal model processes and generates multiple data types: text, images, audio and video.

N

Fondamental

Neural Network

A neural network is a computing model inspired by the human brain, composed of layers of interconnected nodes that process information to learn patterns.

Tendance

Neural Scaling Laws

Neural scaling laws are predictable mathematical relationships linking an AI model's performance to its size, training data volume and compute budget.

Technique

NLP (Natural Language Processing)

Natural Language Processing

NLP is the field of AI that enables machines to understand, interpret and generate human language.

O

Écosystème

Open Source AI

Open source AI refers to models and tools whose code and/or weights are freely accessible and modifiable.

Technique

Overfitting

Overfitting occurs when an AI model has over-learned the training data and fails to generalize to new data.

P

Application

Physical AI

Physical AI applies artificial intelligence to the physical world: robotics, autonomous vehicles, drones and connected objects.

Technique

Prompt

A prompt is the instruction or question you give an AI to get a response. It's the interface between you and the model.

Technique

Prompt Engineering

Prompt engineering is the art of crafting precise instructions to get the best results from an AI model.

Q

Technique

Quantization

Quantization reduces the precision of numbers in an AI model to make it smaller and faster, with minimal quality loss.

R

Technique

RAG (Retrieval-Augmented Generation)

Retrieval-Augmented Generation

RAG is a technique that connects an LLM to external data sources to generate more accurate and up-to-date answers.

Technique

RLHF (Reinforcement Learning from Human Feedback)

Reinforcement Learning from Human Feedback

RLHF is a training technique that uses human feedback to align an LLM's behavior with user expectations.

S

Modèle

SLM (Small Language Model)

Small Language Model

An SLM is a compact language model optimized to run on local devices with targeted performance on specific tasks.

Écosystème

Sovereign AI

Sovereign AI refers to a country or region's ability to develop and control its own AI technologies, independently from American or Chinese tech giants.

Application

Speech-to-Text

Speech-to-Text converts spoken words into written text, enabling automatic transcription of meetings, podcasts and calls.

Technique

Structured Output

Structured Output forces an LLM to respond in a predefined structured format (JSON, XML, table) instead of free text.

Technique

Synthetic Data

Synthetic data is data artificially generated by algorithms or AI models, designed to reproduce the statistical properties of real data without containing personal information.

Technique

System Prompt

The System Prompt is the invisible instruction that defines an LLM's behavior, role and constraints before any user interaction.

T

Technique

Temperature (AI)

Temperature

Temperature is a parameter that controls an LLM's creativity level: low for precise responses, high for creative ones.

Technique

Test-time Compute

Test-time compute allocates more computation at inference time to improve response quality, instead of just scaling up the model.

Application

Text-to-Image

Text-to-Image refers to generating images from text descriptions using generative AI models.

Technique

Text-to-Music

Text-to-Music is a generative AI technique that transforms a text description into a complete musical composition, including melody, harmony, rhythm and instrumentation.

Application

Text-to-Speech

Text-to-Speech converts written text into spoken voice using AI, with increasingly realistic results.

Application

Text-to-Video

Text-to-Video generates videos from text descriptions using generative AI.

Technique

Token

A token is the basic unit processed by an LLM. It's a piece of word, punctuation or character that the model uses to understand and generate text.

Technique

Tokenizer

A tokenizer is the algorithm that splits text into tokens (elementary units) before it is processed by an LLM.

Modèle

Transformer

Transformer Architecture

The Transformer is the neural network architecture powering all modern LLMs, invented by Google in 2017.

V

Infrastructure

Vector Database

A vector database stores embeddings for semantic search and RAG at scale.

Tendance

Vibe Coding

Vibe Coding is a software development approach where the developer describes what they want in natural language and lets AI write the code, without worrying about implementation details.

Modèle

Vision-Language Model (VLM)

Vision-Language Model

A Vision-Language Model (VLM) is an AI model capable of simultaneously understanding and reasoning about images and text, unifying visual perception and language understanding.

W

Modèle

World Model

A World Model is an AI model that builds an internal representation of the physical world to understand and predict events.

Z

Technique

Zero-shot

Zero-shot Learning

Zero-shot enables a model to accomplish a task without any prior examples, solely from the instruction.