The AI Glossary You Didn’t Know You Needed

If AI conversations feel like alphabet soup — AGI here, RAG there, open models, transformers, embeddings — you’re not alone. The technology is moving faster than our ability to name it. Yet understanding the language of AI is no longer optional. It’s the new literacy of the intelligence economy.

Below is a curated set of terms that dominate AI discussions today. Not the full encyclopedia — just the essentials you actually hear in meetings, podcasts and late-night debates.

Core Concepts

AI (Artificial Intelligence)
Systems that perform tasks requiring human-like reasoning.

ML (Machine Learning)
A branch of AI where models learn from data rather than rules.

Deep Learning
Machine learning powered by multi-layered neural networks.

LLM (Large Language Model)
The engines behind modern AI assistants like GPT, Claude, Gemini, Llama.

Levels of Intelligence

ANI – Artificial Narrow Intelligence
AI specialized in a single task.

AGI – Artificial General Intelligence
AI with human-like general cognitive abilities.

ASI – Artificial Superintelligence
A hypothetical intelligence far beyond human capability.

Model Types & Architectures

Open Models
Open-source systems like Llama or Mistral.

Closed Models
Proprietary models such as GPT-5.1 or Gemini.

Foundation Models
Large, general-purpose models that power many applications.

Multimodal Models
Systems that understand text, images, audio and video in one model.

Transformer Architecture
The breakthrough architecture enabling modern AI capabilities.

How AI Thinks

Token
A chunk of text (word or subword) processed by the model.

Embedding
A numerical “map” of meaning that allows AI to compare concepts.

Inference
The moment the AI generates an output.

Hallucination
When an AI confidently generates something false.

Training & Adaptation

Fine-Tuning
Training a pre-built model for a specific task or domain.

RAG (Retrieval Augmented Generation)
AI that retrieves external information before responding.

Red Teaming
Stress-testing AI for vulnerabilities, bias or harmful outputs.

Infrastructure & Compute

Compute
The raw processing power needed for training and inference.

GPU / TPU
Chips optimized for AI workloads.

Latency
The delay between user input and model response.

Scalability
The ability to support more users, data or tasks without breaking.

Ethics & Safety

Alignment
Ensuring AI systems behave as intended.

Guardrails
Safety systems that prevent misuse or harmful output.

Bias
Built-in unfairness caused by skewed data or design.

Explainability
Understanding how a model arrived at a decision.

Product & Workflow

AI Agent
Software that can take actions, not just generate text.

Copilot
An AI assistant built into apps you already use.

API
The digital gateway that lets developers plug AI into products.

Orchestration
Managing multiple models, tools and workflows in a coordinated way.

Why This Matters

AI is no longer a niche field for engineers. It’s becoming the operating system of modern work. Knowing the vocabulary helps you make better decisions, ask sharper questions, and see where the world is heading — before everyone else does.

Welcome to the new fluency.