AI Glossary

Key jargon

Term	Plain-language meaning
Term	A word or phrase with a defined meaning in a field.
Definition	A concise statement of what a term means in this curriculum.
Alias	An alternative name or abbreviation pointing to the same concept.
Scope note	A clarification limiting how a definition should be interpreted.

Key concepts

Definitions should explain distinctions that change design or evaluation decisions.
Return to the lesson context because one term may have different meanings across research, products, and organizations.

Concept map

flowchart LR
    A["Encounter unfamiliar term"] --> B["Read plain definition"]
    B --> C["Follow lesson context"]
    C --> D["Apply distinction in practice"]

Term	Meaning in this curriculum
Agent	System in which a model selects some actions or next steps from observations and available tools
AI harness	Software/policy around models for context, tools, state, permissions, validation, recovery, observability, and evals
CAG	Cache-augmented generation; reuse of prepared context/cache where supported, not a universal replacement for retrieval
Context	Information available to one model invocation
Embedding	Vector representation optimized so some relationships are useful geometrically
Evaluation	Structured measurement of system behavior against defined tasks and criteria
Fine-tuning	Additional optimization of model parameters for data or behavior objectives
Foundation model	Broadly trained model adaptable to many downstream tasks
Grounding	Connecting output to supplied evidence or an external source of truth
Hallucination	Unsupported or false generated content presented as grounded
Inference	Running a trained model to produce predictions/output
LLM	Large language model, usually trained to model token sequences
Memory	Persisted state or information available beyond one invocation; always qualify its type
MCP	Model Context Protocol, an interoperability protocol for hosts/clients/servers and capabilities
Multimodal	Processing or generating more than one modality such as text, image, audio, or video
Prompt	Instructions and/or data supplied to a model; only part of the context
RAG	Retrieval-augmented generation; retrieved material is supplied to a generator
Reranker	Component that reorders retrieval candidates using an additional relevance model or rule
Structured output	Model output constrained to a machine-validated schema
Token	Model/tokenizer-specific unit represented by an integer ID
Tool	External capability exposed through a schema and executed by the host/harness
Transformer	Attention-centered neural architecture introduced for sequence transduction and widely extended since
Workflow	Code-defined sequence or graph of steps and transitions