Glossary
See a word you don't understand? Look it up here. Every entry gives you: a one-sentence explanation + an analogy + when you'll run into it.
Core Concepts
Generative AI
AI that can create new content (text, images, audio, video), rather than just sorting things or making predictions. Analogy: It doesn't "find" the answer for you — it "writes" the answer for you.
Large Language Model (LLM)
A model that has read a huge amount of text and is good at predicting "the next word." It's what powers ChatGPT, Claude, and Gemini. Analogy: Someone who's read the whole internet and is great at finishing your sentences. → Learn more
Token
The smallest unit of text a model works with. In English, a token is roughly three-quarters of a word; in Chinese, one character is about 1–2 tokens. You'll run into it: Every length limit and every bit of pricing is counted in tokens.
Context Window
The maximum amount of content a model can "see" at once, including your question and its answer. You'll run into it: When a conversation gets too long and it starts forgetting earlier parts → start a new chat or paste the key points again.
Parameters
The number of internal "dials" in a model, often used to describe how big a model is (e.g., 7B, 70B). Note: More parameters doesn't always mean more useful. Don't get hung up on the numbers.
Knowledge Cutoff
The point in time where a model's training data ends. It doesn't know about anything that happened after that. You'll run into it: When it can't answer questions about recent news → turn on the search feature.
Using AI
Prompt
What you type in to the AI. → How to write a good one
Prompt Engineering
A systematic approach to designing prompts so you get consistent, high-quality results. In plain terms: Learning how to say what you mean clearly.
Hallucination
When AI produces content that sounds smooth and confident but is wrong. Analogy: Someone who makes up stories just to avoid an awkward silence. → How to spot and fact-check it
Temperature
A setting that controls how random the output is. Low = cautious and consistent; high = wide-ranging and creative.
System Prompt
Rules and a role set up before the conversation starts, which the user usually can't see. You'll run into it: The "Instructions" field in custom GPTs or Claude Projects is exactly this.
Few-shot
Giving a few examples in your prompt so the AI can copy the style you want. This is one of the techniques with the biggest payoff for the least effort.
Chain of Thought (CoT)
Asking the AI to reason step by step before giving its conclusion, which makes it more accurate on complex tasks.
Reasoning Model
A model that "thinks a bit longer" internally before answering. Good for math, logic, and debugging. The trade-off: Slower, and usually more expensive.
Multimodal
A model that can handle text, images, audio, and video together. You'll run into it: When you drop a screenshot or PDF straight into an AI and have it read it.
Advanced and Technical
RAG (Retrieval-Augmented Generation)
Having the AI first search a specific set of documents, then answer based on what it found. Analogy: An open-book exam. The answers come from the book you gave it, not from its memory. The benefit: It dramatically reduces hallucinations, and you can trace where the answers came from. → How to set it up
Vector Database
The kind of database behind RAG that lets you "search by meaning." Analogy: Instead of matching keywords, it matches on "is the meaning similar?"
AI Agent
An AI you give a goal to, and it breaks down the steps, uses tools, and checks the results on its own. Analogy: Going from "an assistant who answers one question at a time" to "a coworker you can hand off a whole task to."
MCP (Model Context Protocol)
An open standard that lets AI connect safely to outside tools and data sources. Analogy: The USB plug standard of the AI world.
Fine-tuning
Training a model further on your own data so it better fits a specific need. Note: Most of the time, what you actually need is RAG or a good prompt, not fine-tuning.
API
A way for different programs to talk to each other. It's the foundation of connecting tools together for automation.
Open-weight Model
A model whose weights are published, so you can download and run it yourself. The benefit: Your data never leaves your own machine, which gives you the most privacy.
Local / On-device
A model that runs directly on your own computer or phone, without going through the cloud.
Guardrails
Safety mechanisms that limit what AI can output, to keep it from producing harmful or inappropriate content.
Alignment
The field of research and engineering focused on making AI behave in line with human intentions and values.
Tools and Applications
Vibe Coding
A way of building software where you describe what you need in plain language and let AI write the code. → Get started
Custom Assistants (GPTs / Projects / Gems)
AI assistants set up ahead of time with a role, rules, and reference materials, so you don't have to explain everything again each time.
Deep Research
The AI spends anywhere from a few minutes to tens of minutes searching multiple sources on its own, then produces a report with citations.
Prompt Injection
Malicious content hidden in a web page or file that tricks the AI into following instructions you didn't intend. You'll run into it: Be careful when you have AI read web pages from unknown sources.
De-identification
Replacing personal identifying information in data with codes or placeholders before using it. → How to do it
Feel free to email us, and we'll keep adding more.