AI Glossary
Clear explanations of RAG, agents, MoE, LoRA and other AI terms: a one-line definition, a plain-language explanation, an example and related tools.
80 terms
AI Agent
An AI system that perceives a goal, plans steps, calls tools and executes tasks.
Alignment
Research and engineering to make AI systems behave according to human intent, values and safety constraints.
Attention Mechanism
A computation that lets a model dynamically weigh how relevant other positions are when processing one position.
Computer Use
AI controlling a computer interface through screenshots, clicks and typing to complete tasks.
Context Rot
The phenomenon where a model uses key information worse as the context grows longer.
Context Window
The maximum number of tokens a model can read and remember at once, including input and output.
Deep Research
AI autonomously running multi-round search, reading, cross-checking and producing a sourced research report.
Diffusion Model
A generative model that turns random noise into images, video or audio through iterative denoising.
Graph RAG
Building a graph of entities and relationships from documents and combining graph retrieval with LLMs for complex questions.
Hallucination
When a model produces plausible but factually wrong or unsupported content.
Large Language Model
A model trained on massive text to understand and generate natural language.
Model Context Protocol
An open standard that lets AI applications connect to external tools and data sources.
Open-Weight Model
A model whose weights can be downloaded, deployed locally and modified.
PII (Personal Identifiable Information)
Information that can identify a specific person alone or in combination, such as name, phone, ID number and address.
Prompt Engineering
Designing instructions, examples and constraints so a model reliably produces the desired output.
Prompt Injection
An attack that hides malicious instructions in user input or external content to hijack model behavior.
Reasoning Model
A model optimized for complex problem solving by generating a longer internal reasoning process before answering.
Retrieval-Augmented Generation
Retrieve relevant passages from a knowledge base, then have the model answer from them.
Text-to-Video
Generating video clips from text descriptions.
Token
The smallest unit of text a model processes, often a word, subword or character group.
Transformer
A neural network architecture built around self-attention and parallel training, the foundation of modern large models.
Vibe Coding
Programming by describing intent in natural language and letting AI generate and modify code.
AI Upscaling
Using models to increase image resolution and add detail.
Agent Memory
A mechanism for an agent to store and retrieve user preferences, past tasks and key facts across sessions.
Agentic Workflow
Breaking a task into steps planned by AI, executed by tools and reviewed by humans.
Artificial General Intelligence
General AI that matches or exceeds humans across a wide range of tasks and can transfer learning.
Automatic Speech Recognition
Converting speech into text.
Benchmark
A standardized dataset and metric for comparing model capabilities.
Brain-Computer Interface
Technology that reads or stimulates brain signals to interact with external devices.
Chain of Thought
Prompting a model to write intermediate reasoning steps before concluding.
Chunking
Splitting long documents into smaller passages for retrieval and embedding, a key step in RAG quality.
ControlNet
An add-on that gives diffusion models structural control from pose, edges, depth and more.
Digital Human
An AI-generated or driven virtual person that can speak and move.
Direct Preference Optimization
An alignment method that optimizes directly on preference data without a separate reward model.
Embedding
Converting text, images or other content into vectors whose similarity can be computed.
Embodied AI
AI that perceives, decides and acts through a physical body.
Few-Shot
Providing a small number of input-output examples in the prompt so the model can imitate the pattern.
Fine-tuning
Continue training a model on task data to better match a style or task.
Function Calling
A model selects and calls external functions based on user intent, then uses the result in its answer.
Guardrails
Rules and systems that constrain model inputs, outputs and actions.
Hybrid Search
Combining keyword search and vector search, then fusing the two result sets.
Image-to-Video
Generating motion video starting from a still image.
In-Context Learning
A model learning a task temporarily from instructions and examples in the prompt without updating its weights.
Inference
The computation a trained model performs to produce output from an input.
Inpainting
Repainting only a masked region of an image while keeping the rest.
Instruction Tuning
Continued training on instruction-response data so a model follows human instructions better.
Interpretability
Research into how models internally represent features and make decisions.
Jailbreak
An attack that bypasses model safety limits through role-play, encoding or multi-turn manipulation.
KV Cache
Caching keys and values of processed tokens during inference so history is not recomputed for every new token.
Latency
The time from sending a request to receiving the model response.
Latent Space
A compressed high-dimensional representation space where diffusion often operates.
Low-Rank Adaptation
Freezing the base model and training a small low-rank adapter for a new task.
Mixture of Experts
Splitting a model into expert subnetworks and activating only a subset per input.
Model Collapse
Training models repeatedly on their own generated data, causing diversity loss and error accumulation.
Model Distillation
Training a smaller student model on a larger teacher model's outputs to compress capability.
Model Evaluation
Designing test sets and scoring rules to decide whether a model works for a specific task.
Multi-Agent
Multiple agents collaborating with divided roles to complete complex tasks.
Multimodal
A model that can process or generate multiple modalities such as text, images, audio and video.
Next-Token Prediction
The training objective of predicting the next token given the preceding text.
Optical Character Recognition
Converting text in images or scans into editable text.
Outpainting
Generating new content beyond the original image boundary.
Positional Encoding
A mechanism that injects word-order information into a Transformer, since self-attention alone is order-agnostic.
Quantization
Representing model weights at lower precision to reduce memory and inference cost.
RAG vs Fine-tuning
RAG handles changing knowledge; fine-tuning handles behavior and style customization.
Red Teaming
Proactively simulating attackers to find security weaknesses in a model or AI system before release.
Reinforcement Learning from Human Feedback
Train a reward model from human preferences, then optimize model outputs with reinforcement learning.
Reranking
Re-scoring and reordering retrieved passages with a stronger model after initial retrieval.
Speculative Decoding
A small model drafts several tokens and a large model verifies them in parallel to speed up generation.
Synthetic Data
Artificially generated data used for training or evaluation.
Temperature
A parameter controlling output randomness; higher values produce more diverse results.
Test-Time Compute
Spending more computation at inference time—longer thinking, multiple samples or search—to improve answer quality.
Text-to-Speech
Converting text into natural-sounding speech.
Tool Use
A model using external tools such as search, code execution or APIs during reasoning.
Top-p Sampling
Sample only from the smallest set of tokens whose cumulative probability reaches p.
Vector Database
A database built to store vectors and retrieve them by similarity.
Voice Agent
A voice AI system that listens, speaks, handles interruptions and executes tasks in real time.
Voice Cloning
Replicating a person's timbre and speaking style from a small sample.
Watermarking
Embedding detectable markers in AI-generated text, images or audio for provenance.
World Model
An internal representation that learns how an environment works and predicts the consequences of actions.
Zero-Shot
A model performing a task without any examples of that task in the prompt.