Free reference
AI Glossary
AI terms in plain English. Someone said a word at you in a meeting — look it up and get on with your day.
Browse by category
- AI Foundations25
- Generative AI & LLMs25
- Prompting25
- RAG & Search25
- Agents & Agentic AI25
- Model Training & Adaptation25
- Evaluation & Quality25
- AI Safety & Responsible AI25
- Security & Privacy25
- Enterprise AI25
- APIs & Integration25
- MCP & Tool Protocols25
- Coding & Software AI25
- Data & Databases25
- Multimodal AI25
- Infrastructure & Compute25
- Model Serving & Economics25
- Automation & Workflows25
- Advanced Concepts25
- Governance, Law & Compliance25
AI Foundations
Generative AI & LLMs
- Attention
- Chat model
- Completion
- Context window
- Deterministic output
- Emergent capability
- Foundation model
- Generative AIGenAI
- Instruction tuning
- Large language modelLLM
- Max tokens
- Next-token prediction
- Post-training
- Pretraining
- Prompt
- Sampling
- Self-attention
- Stop sequence
- System prompt
- Temperature
- Token
- Tokenization
- Tokenizer
- Top-pNucleus sampling
- Transformer
Prompting
- Chain-of-thought prompting
- Context engineering
- Critique prompt
- Decomposition
- Delimiters
- Few-shot prompting
- Grounding prompt
- Indirect prompt injection
- Jailbreak
- Negative instruction
- One-shot prompting
- Output constraint
- Persona prompt
- Prompt chaining
- Prompt engineering
- Prompt injection
- Prompt library
- Prompt optimization
- Prompt template
- Prompt variable
- Prompt versioning
- Reflection prompt
- Role prompting
- Structured prompt
- Zero-shot prompting
RAG & Search
- BM25
- Chunk overlap
- Chunk size
- Chunking
- Citation grounding
- Cosine similarity
- Dense retrieval
- Embedding
- Embedding model
- Hybrid search
- Keyword search
- Knowledge base
- Metadata filtering
- Query expansion
- Query rewriting
- Re-ranking
- Retrieval
- Retrieval quality
- Retrieval-augmented generationRAG
- Retriever
- Semantic search
- Similarity score
- Sparse retrieval
- Vector database
- Vector search
Agents & Agentic AI
- Agent handoff
- Agent loop
- Agent state
- Agentic AI
- Agentic workflow
- AI agent
- Approval gate
- Autonomous agent
- Executor
- Function calling
- Guardrail
- Human-in-the-loopHITL
- Human-on-the-loopHOTL
- Long-term memory
- Memory
- Multi-agent system
- Orchestrator
- Planner
- Short-term memory
- Subagent
- Task queue
- Tool calling
- Tool use
- Workflow
- Working memory
Model Training & Adaptation
- Backpropagation
- Catastrophic forgetting
- Checkpoint
- Continual learning
- Curriculum learning
- Data augmentation
- Direct preference optimizationDPO
- Distillation
- Domain adaptation
- Fine-tuning
- Gradient
- Gradient descent
- Knowledge distillation
- Optimizer
- Preference data
- Reinforcement learningRL
- Reinforcement learning from AI feedbackRLAIF
- Reinforcement learning from human feedbackRLHF
- Reward model
- Student model
- Supervised fine-tuningSFT
- Synthetic data
- Teacher model
- Training run
- Transfer learning
Evaluation & Quality
AI Safety & Responsible AI
- Accountability
- Adversarial example
- Adversarial testing
- AI safety
- Alignment
- Auditability
- Bias
- Content moderation
- Data leakage
- Data privacy
- Dual-use
- ExplainabilityXAI
- Fairness
- Interpretability
- Misuse
- Model alignment
- Model card
- Model leakage
- Red teaming
- Refusal
- Responsible AI
- Safety filter
- Safety policy
- System card
- Transparency
Security & Privacy
- Abuse detection
- AI security
- API key
- Attack surface
- AuthenticationAuthN
- AuthorizationAuthZ
- Backdoor
- Data exfiltration
- Data poisoning
- Data residency
- Encryption
- Isolation
- Least privilege
- Membership inference
- Model extraction
- Model poisoning
- PHIProtected health information
- PIIPersonally identifiable information
- Prompt leakage
- Rate limiting
- Sandbox
- Secrets management
- Supply-chain risk
- Threat model
- Zero trust
Enterprise AI
- AI adoption
- AI fluency
- AI governance
- AI literacy
- AI maturity
- AI operating model
- AI policy
- AI readiness
- AI roadmap
- AI strategy
- AI transformation
- AI use case
- Center of excellenceCoE
- Change management
- Enterprise AI
- Model governance
- Pilot
- Production deployment
- Proof of conceptPoC
- Reskilling
- Return on investmentROI
- Shadow AI
- Total cost of ownershipTCO
- Upskilling
- Workflow redesign
APIs & Integration
- APIApplication programming interface
- API gateway
- Authentication token
- Bearer token
- Client library
- Concurrency
- Endpoint
- Exponential backoff
- Idempotency
- JSONJavaScript Object Notation
- Latency
- Middleware
- OAuthOpen Authorization
- Payload
- Request
- Response
- REST APIRepresentational State Transfer API
- Retry
- Schema
- SDKSoftware development kit
- Streaming
- Structured output
- Throughput
- Timeout
- Webhook
MCP & Tool Protocols
- Capability discovery
- Connector
- Context provider
- Context source
- Integration layer
- Interoperability
- JSON Schema
- Local tool
- MCP client
- MCP prompt
- MCP resource
- MCP server
- MCP tool
- Model Context ProtocolMCP
- Protocol
- Remote tool
- Resource URI
- Tool approval
- Tool error
- Tool permission
- Tool registry
- Tool result
- Tool routing
- Tool schema
- Tool selection
Coding & Software AI
- AI coding assistant
- AI pair programming
- Branch
- Build
- CI/CDContinuous integration and continuous delivery/deployment
- Code completion
- Code generation
- Code review
- Commit
- Debugging
- Dependency
- Deployment
- Environment variable
- Framework
- Git
- Integration test
- Merge
- Package
- Pull requestPR
- Refactoring
- RepositoryRepo
- Test generation
- Unit test
- Version control
- Vibe coding
Data & Databases
- Data annotation
- Data catalog
- Data cleaning
- Data governance
- Data ingestion
- Data labeling
- Data lake
- Data lakehouse
- Data lineage
- Data pipeline
- Data provenance
- Data quality
- Data transformation
- Data warehouse
- Database
- Document database
- ELTExtract, load, transform
- ETLExtract, transform, load
- Key-value store
- NoSQLNot only SQL
- Relational database
- Semi-structured data
- SQLStructured Query Language
- Structured data
- Unstructured data
Multimodal AI
- Audio model
- Automatic speech recognitionASR
- Cross-modal retrieval
- Diffusion model
- Image editing
- Image generation
- Image understanding
- Image-to-image
- Inpainting
- Latent diffusion
- Modality
- Multimodal AI
- Multimodal prompt
- OCROptical character recognition
- Outpainting
- Speech recognition
- Speech synthesis
- Text-to-image
- Text-to-speechTTS
- Text-to-video
- Video generation
- Vision model
- Vision-language modelVLM
- Visual question answeringVQA
- Voice cloning
Infrastructure & Compute
- Accelerator
- Autoscaling
- Caching
- Cloud computing
- Cloud provider
- Cluster
- Cold start
- Compute
- Container
- CPUCentral processing unit
- Distributed training
- GPUGraphics processing unit
- Inference compute
- KubernetesK8s
- Memory bandwidth
- Model compression
- Node
- Parallelism
- Quantization
- Server
- Serverless
- Sharding
- TPUTensor processing unit
- Training compute
- VRAMVideo random-access memory
Model Serving & Economics
- Closed model
- Compute efficiency
- Cost per million tokens
- Cost per request
- Expert model
- Fallback model
- Frontier model
- Hosted model
- Inference cost
- Inference endpoint
- Input tokens
- Latency budget
- Mixture of expertsMoE
- Model router
- Model serving
- Observability
- Open-source model
- Open-weight model
- Output tokens
- Proprietary model
- Routing model
- Service-level agreementSLA
- Small language modelSLM
- Token cost
- Uptime
Automation & Workflows
- Action
- AI automation
- Automation
- Branching logic
- Business process automationBPA
- Condition
- Cron job
- Decision node
- Event-driven workflow
- Exception handling
- Intelligent automation
- Job
- Loop
- Low-code
- Manual review
- No-code
- Parallel step
- Queue
- Robotic process automationRPA
- Rule-based automation
- Scheduled workflow
- Trigger
- Workflow automation
- Workflow builder
- Workflow orchestration
Advanced Concepts
- Artificial general intelligenceAGI
- Artificial superintelligenceASI
- Capability
- Capability overhang
- Chain of thoughtCoT
- Emergence
- General-purpose AIGPAI
- Inference-time scaling
- Latent representation
- Latent space
- Long-horizon task
- Mechanistic interpretability
- Narrow AI
- Parameter count
- Planning horizon
- Reasoning model
- Reasoning trace
- Representation learning
- Scaling law
- Search over reasoning
- Self-consistency
- Sparsity
- Test-time compute
- Tree of thoughtsToT
- World model
Governance, Law & Compliance
- AI compliance
- AI inventory
- AI regulation
- Algorithmic impact assessmentAIA
- Audit trail
- Automated decision-makingADM
- Conformity assessment
- Consent
- Data minimization
- Data processing agreementDPA
- Data protection impact assessmentDPIA
- Documentation requirement
- Governance committee
- High-risk AI
- Human oversight
- Impact assessment
- Model inventory
- Prohibited AI practice
- Purpose limitation
- Recordkeeping
- Right to explanation
- Risk assessment
- Risk tier
- Third-party risk
- Vendor risk
All terms, A to Z
A
- A/B test
- Abuse detection
- Accelerator
- Accountability
- Accuracy
- Action
- Adversarial example
- Adversarial testing
- Agent handoff
- Agent loop
- Agent state
- Agentic AI
- Agentic workflow
- AI adoption
- AI agent
- AI automation
- AI coding assistant
- AI compliance
- AI fluency
- AI governance
- AI inventory
- AI literacy
- AI maturity
- AI operating model
- AI pair programming
- AI policy
- AI readiness
- AI regulation
- AI roadmap
- AI safety
- AI security
- AI strategy
- AI transformation
- AI use case
- Algorithm
- Algorithmic impact assessment
- Alignment
- API
- API gateway
- API key
- Approval gate
- Architecture
- Artificial general intelligence
- Artificial intelligence
- Artificial superintelligence
- Attack surface
- Attention
- Audio model
- Audit trail
- Auditability
- Authentication
- Authentication token
- Authorization
- Automated decision-making
- Automatic speech recognition
- Automation
- Autonomous agent
- Autoscaling
B
C
- Caching
- Capability
- Capability discovery
- Capability overhang
- Catastrophic forgetting
- Center of excellence
- Chain of thought
- Chain-of-thought prompting
- Change management
- Chat model
- Checkpoint
- Chunk overlap
- Chunk size
- Chunking
- CI/CD
- Citation grounding
- Client library
- Closed model
- Cloud computing
- Cloud provider
- Cluster
- Code completion
- Code generation
- Code review
- Cold start
- Commit
- Completion
- Compute
- Compute efficiency
- Concurrency
- Condition
- Conformity assessment
- Confusion matrix
- Connector
- Consent
- Container
- Content moderation
- Context engineering
- Context provider
- Context source
- Context window
- Continual learning
- Cosine similarity
- Cost per million tokens
- Cost per request
- CPU
- Critique prompt
- Cron job
- Cross-modal retrieval
- Curriculum learning
D
- Data annotation
- Data augmentation
- Data catalog
- Data cleaning
- Data exfiltration
- Data governance
- Data ingestion
- Data labeling
- Data lake
- Data lakehouse
- Data leakage
- Data lineage
- Data minimization
- Data pipeline
- Data poisoning
- Data privacy
- Data processing agreement
- Data protection impact assessment
- Data provenance
- Data quality
- Data residency
- Data transformation
- Data warehouse
- Database
- Dataset
- Debugging
- Decision node
- Decomposition
- Deep learning
- Delimiters
- Dense retrieval
- Dependency
- Deployment
- Deterministic output
- Diffusion model
- Direct preference optimization
- Distillation
- Distributed training
- Document database
- Documentation requirement
- Domain adaptation
- Dual-use
E
F
G
H
I
- Idempotency
- Image editing
- Image generation
- Image understanding
- Image-to-image
- Impact assessment
- Indirect prompt injection
- Inference
- Inference compute
- Inference cost
- Inference endpoint
- Inference-time scaling
- Inpainting
- Input tokens
- Instruction tuning
- Integration layer
- Integration test
- Intelligent automation
- Interoperability
- Interpretability
- Isolation
L
M
- Machine learning
- Manual review
- Max tokens
- MCP client
- MCP prompt
- MCP resource
- MCP server
- MCP tool
- Mechanistic interpretability
- Membership inference
- Memory
- Memory bandwidth
- Merge
- Metadata filtering
- Middleware
- Misuse
- Mixture of experts
- Modality
- Model
- Model alignment
- Model card
- Model compression
- Model Context Protocol
- Model extraction
- Model governance
- Model inventory
- Model leakage
- Model poisoning
- Model router
- Model serving
- Multi-agent system
- Multimodal AI
- Multimodal prompt
O
P
- Package
- Parallel step
- Parallelism
- Parameter
- Parameter count
- Pass rate
- Payload
- Persona prompt
- PHI
- PII
- Pilot
- Planner
- Planning horizon
- Post-training
- Precision
- Preference data
- Pretraining
- Production deployment
- Prohibited AI practice
- Prompt
- Prompt chaining
- Prompt engineering
- Prompt injection
- Prompt leakage
- Prompt library
- Prompt optimization
- Prompt template
- Prompt variable
- Prompt versioning
- Proof of concept
- Proprietary model
- Protocol
- Pull request
- Purpose limitation
R
- Rate limiting
- Re-ranking
- Reasoning model
- Reasoning trace
- Recall
- Recordkeeping
- Red teaming
- Refactoring
- Reflection prompt
- Refusal
- Regression test
- Reinforcement learning
- Reinforcement learning from AI feedback
- Reinforcement learning from human feedback
- Relational database
- Relevance
- Remote tool
- Repository
- Representation learning
- Request
- Reskilling
- Resource URI
- Response
- Responsible AI
- REST API
- Retrieval
- Retrieval quality
- Retrieval-augmented generation
- Retriever
- Retry
- Return on investment
- Reward model
- Right to explanation
- Risk assessment
- Risk tier
- Robotic process automation
- Robustness
- Role prompting
- Routing model
- Rubric
- Rule-based automation
S
- Safety filter
- Safety policy
- Sampling
- Sandbox
- Scaling law
- Scheduled workflow
- Schema
- SDK
- Search over reasoning
- Secrets management
- Self-attention
- Self-consistency
- Semantic search
- Semi-structured data
- Server
- Serverless
- Service-level agreement
- Shadow AI
- Sharding
- Short-term memory
- Similarity score
- Small language model
- Sparse retrieval
- Sparsity
- Speech recognition
- Speech synthesis
- SQL
- Stop sequence
- Streaming
- Structured data
- Structured output
- Structured prompt
- Student model
- Subagent
- Supervised fine-tuning
- Supply-chain risk
- Synthetic data
- System card
- System prompt
T
- Task queue
- Teacher model
- Temperature
- Test generation
- Test set
- Test-time compute
- Text-to-image
- Text-to-speech
- Text-to-video
- Third-party risk
- Threat model
- Throughput
- Timeout
- Token
- Token cost
- Tokenization
- Tokenizer
- Tool approval
- Tool calling
- Tool error
- Tool permission
- Tool registry
- Tool result
- Tool routing
- Tool schema
- Tool selection
- Tool use
- Top-p
- Total cost of ownership
- TPU
- Training
- Training compute
- Training run
- Transfer learning
- Transformer
- Transparency
- Tree of thoughts
- Trigger