AI & LLM
Glossary

Plain-English definitions of the terms that actually matter — no jargon for jargon's sake. Each entry links to a full module if you want to go deeper.

See also: AI Model & LLM Pricing — compare what the major models cost.

• AI AgentAn AI system that can take actions and use tools to accomplish a goal, not just chat.
• AttentionThe mechanism that lets a model weigh which tokens matter for each prediction.
• Chain-of-Thought (CoT)Prompting a model to reason step by step before answering.
• Context WindowThe maximum amount of text (in tokens) a model can consider at once.
• DistillationTraining a smaller 'student' model to mimic a larger 'teacher' model.
• EmbeddingA numeric vector that captures the meaning of text so machines can compare it.
• Few-Shot PromptingShowing a model 2–5 examples of the pattern you want before your real input.
• Fine-TuningFurther-training a base model on specialized data to specialize its behavior.
• HallucinationWhen a model produces fluent output that is factually wrong or made up.
• InferenceRunning a trained model to generate output — what happens on every prompt.
• Large Language Model (LLM)An AI model trained on vast amounts of text to predict and generate language.
• Mixture of Experts (MoE)An architecture that activates only part of the model for each token, saving compute.
• Model Context Protocol (MCP)An open standard for connecting AI models to tools and data sources.
• Multimodal AIModels that handle more than text — images, audio, or video alongside language.
• Open-Weights ModelA model whose trained parameters are publicly released to download and run.
• ParametersThe internal numbers a model learns during training; roughly, its 'knobs'.
• Prompt EngineeringThe craft of writing inputs that get the best output from a model.
• Prompt InjectionA security attack where malicious text hijacks a model's instructions.
• QuantizationCompressing a model by storing its numbers at lower precision to cut cost.
• Reasoning ModelA model trained to deliberate at length before answering hard problems.
• Retrieval-Augmented Generation (RAG)Giving a model relevant external documents at query time so it answers from facts.
• System PromptHigh-level instructions that set a model's role, rules, and behavior for a session.
• TemperatureA setting that controls how random vs. deterministic a model's output is.
• TokenThe basic unit of text an LLM reads and generates — roughly a word-piece.
• Top-k SamplingA sampling setting that keeps only the k most likely next tokens.
• Top-p (Nucleus Sampling)A sampling setting that keeps the smallest set of tokens covering probability p.
• TrainingThe compute-intensive process of teaching a model by adjusting its parameters.
• TransformerThe neural-network architecture behind virtually all modern LLMs.
• Vector DatabaseA database built to store embeddings and find the most similar ones fast.
• Zero-Shot PromptingAsking a model to do a task directly, with no examples.