AI Terminology Demystified
The AI world is full of jargon. Models, tokens, context windows, agents, RAG, fine-tuning. This glossary cuts through the noise. Every term is explained in plain language with a real-world analogy.
Orcha Team
January 11, 2026
The Basics
These are the terms you will encounter in virtually every AI conversation.
LLM (Large Language Model)
The core AI technology behind ChatGPT, Claude, Gemini, and others. A mathematical model trained on vast amounts of text that predicts what comes next in a sequence.
Think of it as: a brain that has read the entire internet and can generate human-like text based on patterns.
Model
A specific version of an LLM with particular capabilities. Different models have different strengths: some are fast and cheap, others are slow and powerful.
Think of it as: car models – a compact car and a sports car are both cars, but they are built for different purposes.
Token
The smallest unit of text that a model processes. Roughly 3–4 characters. “Finance team” = 3 tokens. Models charge per token.
Think of it as: the “word count” that determines cost and limits. 200 pages of text equal roughly 80,000 tokens. A typical prompt uses a few hundred tokens; a long document might be 50,000+.
Context Window
The total amount of text a model can “see” at once – both your input and its output combined. Current models have context windows of roughly 200,000 tokens (about 500 pages).
Think of it as: the AI’s desk. A bigger desk means it can spread out more documents at once, but everything beyond the desk edge is invisible.
Prompt
The text you send to an AI model – your question, instruction, or task description. A good prompt is specific, includes context, and describes the desired output format.
Think of it as: a brief you give to a colleague. The more specific and complete your brief, the better the result.
Models & Providers
The major AI companies and their model families. Knowing who makes what helps you navigate the landscape.
Claude (Anthropic)
A family of AI models. Opus is the most powerful (complex reasoning), Sonnet is the everyday workhorse (fast, balanced), Haiku is the cheapest (quick lookups, simple tasks).
Think of it as: three team members with different seniority levels – you pick the right person for the task at hand.
ChatGPT (OpenAI)
The AI product from OpenAI that kicked off the AI boom in late 2022. Behind ChatGPT is a series of models available via the chat interface and an API. Currently the most widely used AI assistant worldwide.
Think of it as: the model most people tried first. A strong all-rounder with the largest ecosystem of plugins and integrations.
Gemini (Google)
Google’s AI model family. Pro is the most capable, Flash is fast and affordable. Deeply integrated with Google Workspace (Docs, Sheets, Gmail).
Think of it as: the best choice when you live in the Google ecosystem.
Open-Weight Models (Llama, Mistral, etc.)
AI models that can be run on your own servers – for full data control. Currently significantly less capable than the best commercial models, but developing rapidly.
Think of it as: Linux vs. Windows – more control, more setup, strong community.
How AI Works
The key concepts behind what happens when you interact with an AI model.
Training
The process of showing a model billions of examples so it can learn. Training a large model costs millions of dollars and takes months. After training, the model’s knowledge is fixed – it does not learn from your conversations (unless fine-tuned).
Think of it as: school. The model graduates with general knowledge, then applies it to specific tasks.
Inference
When a trained model generates a response to your prompt. This is what happens every time you send a message to ChatGPT or Claude.
Think of it as: the exam. Training was the studying; inference is the test where the model shows what it knows.
Hallucination
When an AI model generates information that sounds plausible but is factually incorrect. Models produce probable text, not verified truth.
Think of it as: a confident colleague who sometimes makes things up without realizing it. Always verify critical facts.
Temperature
A setting that controls how creative vs. predictable a model’s output is. Low temperature (0) = very predictable, same answer every time. High temperature (1) = more creative, more varied.
Think of it as: a dial between “strict accountant” and “creative brainstormer.”
Advanced Concepts
Terms that come up when the conversation moves beyond basic AI usage into building real systems.
Agent
An AI system that can take actions, not just generate text. An agent combines an LLM with tools (web search, file access, APIs) and a goal. It plans, executes steps, observes results, and adjusts until the goal is reached.
Think of it as: the difference between asking someone a question and asking them to complete a task end to end.
RAG (Retrieval-Augmented Generation)
A technique where the AI first searches a knowledge base for relevant documents, then generates a response based on what it found. Solves the problem of AI not knowing your internal data.
Think of it as: giving the AI access to your company’s filing cabinet before it answers your question.
Fine-Tuning
Training a pre-trained model further on your own data to specialize it for your use case. Expensive and complex, but produces models that deeply understand your domain.
Think of it as: sending a generalist to a specialized training program for your industry.
MCP (Model Context Protocol)
A standard that lets AI models connect to external tools and data sources. Instead of copying data into the AI, MCP lets the AI read from your systems directly.
Think of it as: a user getting access to specific tools – with clearly defined permissions and restrictions.
Embedding
A way to convert text into numbers (vectors) that capture meaning. “Invoice” and “bill” would have similar embeddings. Used for semantic search – finding documents by meaning, not just keywords.
Think of it as: a fingerprint for text. Similar meanings have similar fingerprints.
AI in Practice
How these concepts come together when you actually use AI in your day-to-day work.
Prompt Engineering
The skill of writing effective prompts. Includes techniques like giving examples, specifying format, assigning a role, and breaking complex tasks into steps.
Think of it as: a communication skill. The better your brief, the better the result. Finance professionals have a natural advantage here – they are already trained in precision and structured thinking.
Zero-Shot vs. Few-Shot
Zero-shot = asking AI to do something with no examples. Few-shot = including 2–3 examples in your prompt so the model understands the pattern.
Think of it as: showing a new employee a completed report before asking them to write the next one. A few examples dramatically improve quality.
Workflow / Automation
Connecting multiple AI steps into an automated chain that runs without human intervention. Example: email arrives → AI extracts data → validates against ERP → routes for approval.
Think of it as: an assembly line where each station is powered by AI instead of manual work.
Guardrails
Rules and constraints placed on AI systems to prevent undesired behavior. Examples: content filters, output validation, human-in-the-loop requirements for high-stakes decisions.
Think of it as: safety rails. Let the AI work fast, but keep it on track.
You Don’t Need to Know Everything
You do not need to understand every detail of how AI works to use it effectively. But knowing the vocabulary helps you ask better questions, evaluate tools more critically, and have productive conversations with your team about what AI can and cannot do.
Want more guides like this?
Subscribe to our newsletter for new resources delivered to your inbox.