Core AI concepts
Short definitions connected to examples, tools and deeper learning paths.
Foundation model
A model trained at broad scale that can be adapted to many downstream tasks.
Explain →LLM
A large language model that predicts and generates token sequences from context.
Explain →Token
A unit of text processed by a model; it does not map exactly to a word.
Explain →Context window
The maximum amount of information a model can consider in one interaction.
Explain →Prompt
The instruction and context given to a model to guide its output.
Explain →System prompt
Higher-priority instructions that define behavior, constraints and application rules.
Explain →Hallucination
A plausible-looking output that is false or unsupported by the available evidence.
Explain →Grounding
Anchoring a response in verifiable data, documents or external tools.
Explain →Embedding
A numeric vector representing semantic features so items can be compared by similarity.
Explain →Vector database
A system that stores vectors and retrieves nearby items by similarity.
Explain →RAG
Retrieval-augmented generation: retrieve relevant external information and add it to model context before generation.
Explain →Chunk
A fragment of a source created for indexing, retrieval or processing.
Explain →Chunking
The strategy used to split documents while preserving useful meaning and context.
Explain →Reranking
A second-stage process that reorders retrieved candidates with a more precise model or criterion.
Explain →Hybrid search
A retrieval strategy combining lexical and semantic search.
Explain →Fine-tuning
Additional model training on task-specific examples to change behavior or specialize performance.
Explain →Quantization
Reducing numerical precision to lower memory use and speed up inference, with possible quality trade-offs.
Explain →Multimodal
Able to process or generate more than one modality such as text, image, audio or video.
Explain →Reasoning model
A model optimized for tasks that need multi-step inference, planning or verification.
Explain →Inference
Running a trained model to produce an output.
Explain →Latency
The time between a request and a useful response or first output.
Explain →Throughput
The amount of work a system can process per unit of time.
Explain →Prompt caching
Reusing repeated context to reduce cost or latency when a provider supports it.
Explain →Agent
A system that combines a model, state and tools to pursue a goal through controlled actions.
Explain →Tool calling
A mechanism that lets a model request a structured external action.
Explain →MCP
A protocol for connecting models and agents to tools, resources and context through a common interface.
Explain →Prompt injection
Content that tries to override system instructions through data the model reads.
Explain →Evals
Repeatable evaluations used to measure whether a model or system meets defined quality criteria.
Explain →Structured output
Model output constrained to a defined schema instead of free text.
Explain →Model routing
Dynamically selecting a model based on task, cost, risk or latency.
Explain →