AI Glossary

All 1 A B C D E F G H I J K L M N O P Q R S T U V W Z

M

Millette v. OpenAI / Google / NVIDIA

A series of lawsuits related to the collection of content (e.g., videos) for training models, illustrating the expansion of disputes to hardware and platform providers as well.

Minimalist Implementation

An approach of the Czech legislator whereby the adaptation act regulates only the necessary elements and leaves detailed obligations directly to the text of the AI Act, so as not to increase the administrative burden on companies.

Mistral

A European (French) provider of language models, often highlighted in connection with European competitiveness in the field of AI.

Mixture of Experts (MoE)

A model architecture in which only a part of the specialized subnetworks (experts) is activated for each input, which enables larger models at lower computational cost per response.

Model Alignment

An effort to ensure that a model's goals and behavior correspond to human values and intentions, thereby preventing undesirable or harmful conduct.

Model Card

A structured document describing the purpose, capabilities, limitations, and both recommended and inappropriate uses of a model, serving transparency toward users.

Model Collapse

The gradual degradation of the quality of models trained predominantly on content created by another AI, where diversity and fidelity to the original data are lost.

Model Documentation Form

A standardized model documentation form that a GPAI provider must prepare before placing a model on the EU market and which summarizes the information required by Annexes XI and XII of the AI Act.

Model Drift

The gradual deterioration in the accuracy of a deployed model caused by real-world data gradually diverging from the data on which the model was trained.

Model Evaluation (Evals)

Systematic testing of a model's capabilities, safety and reliability using task suites; the basis for deployment decisions and for demonstrating compliance.

Model Extraction

A type of attack in which an attacker mimics the behavior of a target model through repeated queries and creates a functional copy of it.

Model Interpretability

The ability to understand the internal logic of how an AI model works.

Model Inversion

A type of attack that attempts to reconstruct the sensitive input data used during training from a model's outputs.

Model Latency and Throughput

Operational speed metrics: latency expresses the time to the first response, throughput the number of tokens processed per unit of time; they affect usability and cost.

Model License

License terms determining how a model may be used, distributed, and further modified, including any restrictions on commercial use.

Model Parameter

A numerical value inside a model that is set during training; the number of parameters (often in the billions) tends to be a rough indicator of the model's size and capacity.

Model Routing

A technique that automatically routes each query to the most suitable model based on its complexity, cost, or required capabilities.

Model Training

The process by which a model learns to recognize patterns and relationships in data.

Model Weights

A set of learned parameters that determines a model's behavior. Making them available distinguishes open-weight models from closed ones.

Moral Rights

The rights of an author connected with their personality, such as the right to attribution and to the integrity of the work, which cannot be transferred to another person.

Moral Rights in AI Content

The author's moral rights (e.g., the right to attribution of authorship and to the integrity of the work) and their application in connection with content created or modified by AI.

Mosaic / In re LLM Litigation

A collective term for consolidated proceedings bringing together multiple lawsuits against providers of large language models, in which fundamental rulings on fair use are expected.

Multi-Agent Collaboration

An arrangement in which multiple AI agents communicate with one another and share tasks in order to jointly solve more complex assignments.

Multi-Agent System

A system composed of multiple AI agents that cooperate or compete with one another to jointly accomplish tasks or solve complex problems.

Multi-Factor Authentication (MFA)

A security method requiring multiple independent factors to verify identity, for example a password and a one-time code.

Multimodal Model

A model capable of processing and combining multiple types of inputs and outputs, such as text, image, audio, and video.

N

National AI Act Adaptation Law

Czech legislation that supplements the AI Act in areas where the European regulation leaves room for adjustment by Member States.

National AI Strategy

A government document setting out the priorities for the development and regulation of AI in a given country.

National Competent Authority

An authority designated by a Member State to supervise the application of the AI Act on its territory; Member States were to designate them by August 2025.

Neural Network

A computational model inspired by the structure of the brain, composed of interconnected layers of computational units, which forms the basis of modern machine learning.