Token
The unit of text models read, write and bill by.
Using AI
AI Glossary
Plain-English definitions of 232 AI terms, from tokens and context windows to agents and data centers. Every term has its own page with an example.
232 terms/
Start here
New to AI? These six ideas make the rest of the glossary easier to follow.
Context window
How much a model can take into account at once.
Using AI
Prompt
Everything you give a model as input.
Basics
Hallucination
Confident answers that are not true.
Using AI
Retrieval-augmented generation
Looking things up before answering.
Using AI
AI agent
A model that takes steps toward a goal.
Agents & tools
#
- 3D generationImages, video & audio
A
- A2AAgents & tools
- Activation functionModels & architectures
- Agent evalsAgents & tools
- Agent guardrailsAgents & tools
- Agent loopAgents & tools
- Agent memoryAgents & tools
- Agent permissionsAgents & tools
- Agent skillsAgents & tools
- Agentic workflowAgents & tools
- AI acceleratorHardware & infrastructure
- AI agentAgents & tools
- AI avatarImages, video & audio
- AI biasSafety & evaluation
- AI energy useHardware & infrastructure
- AI regulationSafety & evaluation
- AI safetySafety & evaluation
- AlgorithmBasics
- AlignmentSafety & evaluation
- APIUsing AI
- Arena rankingsSafety & evaluation
- Artificial general intelligenceBasics
- Artificial intelligenceBasics
- Attention mechanismModels & architectures
- Autonomous agentAgents & tools
- Autoregressive modelModels & architectures
B
- BackpropagationTraining & data
- Batch APIUsing AI
- Batch sizeTraining & data
- BatchingHardware & infrastructure
- BenchmarkSafety & evaluation
- Benchmark contaminationSafety & evaluation
- BERTModels & architectures
- Browser agentAgents & tools
C
- CapexHardware & infrastructure
- CFG scaleImages, video & audio
- Chain of thoughtUsing AI
- ChatbotBasics
- CheckpointTraining & data
- ChunkingUsing AI
- Cloud AIHardware & infrastructure
- Coding agentAgents & tools
- ComputeTraining & data
- Compute clusterHardware & infrastructure
- Computer useAgents & tools
- Computer visionBasics
- Content provenance (C2PA)Safety & evaluation
- Context engineeringAgents & tools
- Context windowUsing AI
- ControlNetImages, video & audio
- Convolutional neural networkModels & architectures
- CopilotAgents & tools
- CUDAHardware & infrastructure
- Curriculum learningTraining & data
D
- Data centerHardware & infrastructure
- Data center coolingHardware & infrastructure
- Data labelingTraining & data
- Data poisoningSafety & evaluation
- Data privacySafety & evaluation
- DatasetBasics
- DecoderModels & architectures
- Deep learningBasics
- DeepfakeSafety & evaluation
- Denoising strengthImages, video & audio
- Dense modelModels & architectures
- Diffusion modelModels & architectures
- Direct preference optimizationTraining & data
E
- Edge AIHardware & infrastructure
- Embedding modelModels & architectures
- EmbeddingsUsing AI
- EncoderModels & architectures
- Encoder-decoder modelModels & architectures
- EpochTraining & data
- EU AI ActSafety & evaluation
- EvalsSafety & evaluation
- ExplainabilitySafety & evaluation
F
- FairnessSafety & evaluation
- Few-shot promptingUsing AI
- Fine-tuningTraining & data
- FLOPsHardware & infrastructure
- Foundation modelBasics
- Frequency and presence penaltiesUsing AI
- Frontier modelModels & architectures
- Frontier safety frameworkSafety & evaluation
G
- Gaussian splattingImages, video & audio
- Generative adversarial networkModels & architectures
- Generative AIBasics
- GPTModels & architectures
- GPUHardware & infrastructure
- GPU memoryHardware & infrastructure
- Gradient descentTraining & data
- GroundingUsing AI
- GuardrailsSafety & evaluation
H
- HallucinationUsing AI
- HandoffAgents & tools
- High-bandwidth memoryHardware & infrastructure
- Hugging FaceHardware & infrastructure
- Human-in-the-loopAgents & tools
- HyperparameterTraining & data
- HyperscalerHardware & infrastructure
I
- Image-to-imageImages, video & audio
- Image-to-videoImages, video & audio
- In-context learningUsing AI
- InferenceBasics
- Inference costUsing AI
- Inference serverHardware & infrastructure
- InpaintingImages, video & audio
- Instruction tuningTraining & data
- InterconnectHardware & infrastructure
J
- JailbreakSafety & evaluation
K
- Knowledge cutoffUsing AI
- Knowledge distillationTraining & data
L
- Large language modelBasics
- LatencyUsing AI
- Latent diffusionImages, video & audio
- Latent spaceModels & architectures
- LeaderboardSafety & evaluation
- Learning rateTraining & data
- Lip syncImages, video & audio
- LLM-as-a-judgeSafety & evaluation
- Long short-term memoryModels & architectures
- Long-horizon taskAgents & tools
- LoRATraining & data
- LoRA (image models)Images, video & audio
- Loss functionTraining & data
M
- Machine learningBasics
- Max tokensUsing AI
- MCP serverAgents & tools
- Mechanistic interpretabilitySafety & evaluation
- Mixture of expertsModels & architectures
- MMLUSafety & evaluation
- ModelBasics
- Model cardSafety & evaluation
- Model compressionHardware & infrastructure
- Model Context ProtocolAgents & tools
- Model servingHardware & infrastructure
- Multi-agent systemAgents & tools
- Multi-head attentionModels & architectures
- Multimodal AIBasics
- Music generationImages, video & audio
N
- Narrow AIBasics
- Natural language processingBasics
- Negative promptImages, video & audio
- NeRFImages, video & audio
- Neural networkBasics
- NPUHardware & infrastructure
O
- On-device AIHardware & infrastructure
- ONNXHardware & infrastructure
- Open-weight modelModels & architectures
- OrchestrationAgents & tools
- OutpaintingImages, video & audio
- OverfittingTraining & data
P
- ParametersBasics
- PlannerAgents & tools
- PluginsAgents & tools
- Positional encodingModels & architectures
- Power usage effectivenessHardware & infrastructure
- Pre-trainingTraining & data
- PredictionBasics
- ProbabilityBasics
- PromptBasics
- Prompt cachingUsing AI
- Prompt engineeringUsing AI
- Prompt injectionAgents & tools
- PruningHardware & infrastructure
- PyTorchHardware & infrastructure
Q
- QLoRATraining & data
- QuantizationHardware & infrastructure
R
- Rate limitUsing AI
- ReActAgents & tools
- Reasoning modelModels & architectures
- Recurrent neural networkModels & architectures
- Red teamingSafety & evaluation
- RegularizationTraining & data
- Reinforcement learningBasics
- Reinforcement learning from AI feedbackTraining & data
- Reinforcement learning from human feedbackTraining & data
- Reinforcement learning from verifiable rewardsTraining & data
- RerankingUsing AI
- Responsible AISafety & evaluation
- Retrieval-augmented generationUsing AI
- Reward hackingSafety & evaluation
- Reward modelTraining & data
S
- Safety classifierSafety & evaluation
- SamplerImages, video & audio
- Sampling stepsImages, video & audio
- SandboxAgents & tools
- Scaling lawsTraining & data
- SeedImages, video & audio
- Self-attentionModels & architectures
- Self-supervised learningBasics
- Semantic searchUsing AI
- Small language modelModels & architectures
- Speech-to-textImages, video & audio
- State-space modelModels & architectures
- StreamingUsing AI
- Structured outputUsing AI
- Style transferImages, video & audio
- SubagentAgents & tools
- Supervised learningBasics
- SycophancySafety & evaluation
- Synthetic dataTraining & data
- System promptUsing AI
T
- TemperatureUsing AI
- Temporal consistencyImages, video & audio
- Text-to-imageImages, video & audio
- Text-to-speechImages, video & audio
- Text-to-videoImages, video & audio
- TokenUsing AI
- TokenizerUsing AI
- Tool schemaAgents & tools
- Tool useAgents & tools
- Top-pUsing AI
- TPUHardware & infrastructure
- TrainingBasics
- Training dataBasics
- Transfer learningTraining & data
- TransformerModels & architectures
U
- UnderfittingTraining & data
- Unsupervised learningBasics
- UpscalingImages, video & audio
V
- Validation and test setsTraining & data
- Variational autoencoderModels & architectures
- Vector databaseUsing AI
- Vibe codingAgents & tools
- Vision transformerModels & architectures
- Vision-language modelModels & architectures
- Voice cloningImages, video & audio
W
- WatermarkingSafety & evaluation
- WeightsBasics
- Workflow automationAgents & tools
Z
- Zero-shot promptingUsing AI
No terms match.
Try a shorter word or an acronym such as RAG or LLM.
AI Glossary FAQ
How many AI terms does the glossary define?
232 AI terms. Each term has its own page with a plain-English definition, an example and links to related terms.
How do I find a term?
Search terms by name or meaning, filter by topic or browse A to Z, then open the term to read its definition.
Definitions last reviewed .