The problem
A vendor talks about “RAG”, “agents” and “tokens”. You nod along, not quite sure what you are being sold. It is hard to make a good decision when the vocabulary escapes you.
The solution
Here are the thirty terms that come up most, explained simply.
The basics
- Artificial intelligence (AI)
- Software that can do tasks that used to need human judgment: understand text, recognize an image, forecast a trend.
- Generative AI
- AI that produces new content: text, images, sound, code. ChatGPT, Copilot, Gemini and Claude are examples.
- Large language model (LLM)
- The engine behind an assistant like ChatGPT. It learned to predict the next part of a text from huge numbers of examples.
- AI assistant
- The app you chat with. It runs on one or more models.
- Prompt
- The written request you give the AI. A precise prompt (role, context, format, limits) gets a much better answer.
- Token
- The small piece of text a model reads and writes. Providers often bill by the number of tokens.
- Context window
- How much text the model can read at once, prompt and documents included.
- Hallucination
- When AI confidently makes something up. It is why a person always reviews.
Working with AI
- AI agent
- An AI that can use tools to complete a multi-step task: read an email, look up a file, create a task.
- Automation
- A chain of actions that runs on its own when something happens. It does not always need AI.
- Workflow
- The sequence of steps in a process, from request to result.
- Integration
- A link between two apps so they exchange data without retyping.
- API
- The door an app opens to other apps so they can read or write its data.
- RAG
- A method that gives AI your own documents to check before answering. Answers become specific to your business.
- Knowledge base
- Your procedures, answers and documents, organized so the team and AI can look things up.
- Fine-tuning
- Retraining a model on your examples. Rarely needed for a small business.
- Local model
- A model that runs on your own computers or servers.
- Copilot
- AI built into an app to help while you work, like Copilot in Outlook or Word.
- Chatbot
- An assistant that answers questions, for example on a website. It must be able to say “I don’t know”.
- AI receptionist
- A voice assistant that answers the phone, takes messages and books appointments.
Setting rules for AI
- Shadow AI
- Employees using AI tools without approval. The main risk to your data.
- AI use policy
- The document that says which tools are allowed, for what, and what never goes into them.
- Law 25
- Québec’s law on the protection of personal information in the private sector.
- Automated decision
- A decision made only by a system. In Québec, the person must be told.
- Data sovereignty
- Which country’s laws your data falls under.
- Human in the loop
- A person approves AI’s important actions before they take effect.
- Bias
- A tendency of AI to favour certain answers, inherited from its training data.
Getting found
- Search engine optimization (SEO)
- The practices that help your site appear in Google and Bing.
- Generative engine optimization (GEO)
- The practices that help your business get cited in AI assistants’ answers.
- Google Business Profile
- The listing that shows your hours, reviews and contact details in Google and Maps.