AAA game
#Also used as: AAA, triple-A game
An informal label for a high-budget game with large-scale development and marketing; it is not a technical standard or an automatic guarantee of quality.
Glossary
AI glossary
Plain-language explanations of specialist terms used in artificial intelligence, data, security and business adoption.
16 explained terms
Large language model (LLM)
#Also used as: LLM, large language model
A machine-learning model trained on large amounts of text that processes tokens and can generate or analyse language; its responses are not necessarily true or reliable.
Sources
Prompt
#Also used as: model instruction, model input
Text or another input containing an instruction, context or examples that a user or system provides to a model to influence its next output.
Sources
AI agent
#Also used as: agentic system, artificial intelligence agent
A software system that uses a model, tools and control logic to plan and carry out multiple steps towards a specified goal.
Sources
Eval
#Also used as: evaluation, quality test, model evaluation
A repeatable assessment of AI output using data and criteria relevant to a specific task and its risks.
Observability
#Also used as: system observability
The ability to explain a system’s behaviour using logs, metrics, call traces and events, without claiming that these provide a complete view of the model’s internal workings.
Sources
Context window
#Also used as: context length
The token limit that a model or a particular interface can process within a single working context; it is not the system’s persistent memory.
Sources
NPU
#Also used as: neural processing unit
A processor designed to run selected neural-network computations efficiently on a device; by itself, it does not guarantee that all data is processed locally.
Sources
DPA
#Also used as: data processing agreement, personal data processing agreement
A contractual arrangement between a controller and a processor of personal data under Article 28 of the GDPR; its existence alone does not demonstrate full compliance.
Sources
AI literacy
#Also used as: artificial intelligence literacy
The knowledge, skills and understanding needed to deploy and use AI systems in an informed way, taking account of the specific context, opportunities and risks.
Sources
Deployer
#Also used as: AI system deployer
A person or organisation that uses an AI system under its authority in the course of a professional activity; this does not include personal, non-professional use.
Sources
Open-weight model
#Also used as: open weights
A model whose trained weights are available to download or use; this does not automatically mean that the model, data and code meet the requirements of open source.
LLM as a judge
#Also used as: LLM-as-a-judge, model-based evaluator
The use of a large language model to assess outputs against specified criteria; the method can be useful, but it has known biases and is not automatically objective.
Sources
Prompt injection
#Also used as: instruction injection
A manipulative instruction in untrusted input that attempts to change a model’s behaviour or bypass rules set by the system or user.
Sources
WSL 2
#Also used as: Windows Subsystem for Linux 2
The second architecture of Windows Subsystem for Linux, which uses a real Linux kernel in a managed, virtualised Windows environment.
Official pageSources
Zero-day vulnerability
#Also used as: 0-day, zero-day
A vulnerability for which no fix is available at the time it is exploited or disclosed, or for which defenders have not had time to prepare a remedy.
Sources
No glossary term matches this search.