You know that feeling when you walk into a conversation and everyone’s using words you don’t recognize? That’s how most people feel when they start reading about AI.
Prompts. Tokens. Models. Fine-tuning. Context windows. Large language models. It sounds like a foreign language. And honestly, a lot of AI content is written by people who forgot what it was like to not know this stuff.
So here’s your plain-English dictionary. Every term you’ll actually encounter, explained like a normal person would explain it.
The Words That Matter Most
Let’s start with the ones you’ll hear daily.
Prompt A prompt is what you type into the AI. That’s it. Your question. Your instruction. Your request. “Write me a meal plan” is a prompt. “Help me draft an email to my boss” is a prompt.
The word sounds technical. It’s not. It’s just your message to the AI.
A “good prompt” is one that gives the AI enough information to help you properly. A “bad prompt” is one that’s too vague. You already know how to write good ones if you’ve read our article on talking to AI.
Model A model is the AI brain itself. When people say “GPT-4” or “Claude Opus” or “Gemini Pro,” they’re talking about specific models. Think of it like car models. A Honda Civic and a Honda Accord are both Hondas, but they have different capabilities.
Newer models are generally better. More accurate, better at understanding context, better at following instructions. When a company releases a new model, it’s like a software upgrade.
You don’t need to know the technical details of each model. You just need to know that when someone says “use the latest model,” they mean the newest, most capable version.
Token A token is how AI measures the length of text. Not by words. Not by characters. By tokens, which are pieces of words.
Roughly: 1 token equals about 3/4 of a word. So 100 words is about 133 tokens. A full page of text is about 500 to 700 tokens.
Why does this matter? Because AI tools have limits on how many tokens they can process at once. This is called the “context window.” If you give the AI a really long document plus a detailed question, you might hit that limit.
For daily use, you rarely need to worry about tokens. But if you’re working with very long documents or having very long conversations, it’s good to know this limit exists.
Context Window This is how much the AI can “see” at once. Think of it as the AI’s working memory.
If the context window is 100,000 tokens (which is roughly 75,000 words), that means the AI can hold about 75,000 words in its head during your conversation. Your context document, your messages, the AI’s responses, they all fit within this window.
When a conversation gets longer than the window, the AI starts “forgetting” the earliest parts. This is why very long conversations can lose coherence. The AI literally can’t see what you said 50 messages ago.
Practical tip: for long conversations, summarize earlier points or restart with a fresh message that includes the important context.
Hallucination When the AI makes something up and presents it as fact. This is the technical term for “the AI confidently lied to you.”
It’s not doing it on purpose. The way these models work, they predict the most likely next word based on patterns. Sometimes that prediction creates text that sounds true but isn’t. A fake study. A wrong date. A book that doesn’t exist.
The fix is always the same: verify important facts. Especially names, dates, statistics, and quotes.
The Words You’ll Encounter Sometimes
Large Language Model (LLM) This is the type of AI that ChatGPT, Claude, and Gemini all are. A “language model” because it works with language (text). “Large” because it was trained on a massive amount of text data.
When someone says “LLM,” they’re talking about this category of AI. Not a specific product. It’s like saying “smartphone” instead of “iPhone.”
Fine-tuning Taking a general AI model and training it further on specific data to make it better at a particular task.
You probably won’t fine-tune a model yourself. But it’s worth knowing the term because companies use it. “We fine-tuned GPT-4 for customer service” means they took the base model and specialized it.
For personal use, what you’re doing with your context document and consistent interactions is a lighter version of this. You’re not technically fine-tuning the model, but you’re customizing its behavior for your needs.
API (Application Programming Interface) A way for software to talk to other software. When someone says “I’m using the Claude API,” they mean they’re connecting to Claude’s brain from their own program, not through the website.
You don’t need to use an API for personal productivity. The web interface and apps work fine. But if you hear the word, now you know what it means.
System Prompt A special set of instructions given to the AI before your conversation starts. It defines the AI’s behavior, personality, and rules.
On consumer platforms, you interact with this through “Custom Instructions” (ChatGPT) or “Project Instructions” (Claude). It’s where you put your standing preferences. “Always respond in bullet points.” “Use a casual tone.” “You’re my personal productivity assistant.”
This is basically your context document, loaded automatically.
Temperature A setting that controls how creative or predictable the AI’s responses are. Low temperature means more predictable, factual, consistent answers. High temperature means more creative, varied, sometimes surprising answers.
Most users never touch this setting. The default works fine. But if you’re getting responses that feel too robotic, a slightly higher temperature adds personality. If you’re getting responses that feel too wild, lower it.
RAG (Retrieval-Augmented Generation) A fancy way of saying “the AI looks up real information before answering.” Instead of just guessing from its training data, it retrieves relevant documents or data first.
Perplexity is a good example. It searches the web, retrieves information, then generates an answer based on what it found. That’s RAG in action.
For your personal system, when you give the AI your context document and templates, you’re doing a manual version of this. You’re providing the retrieval. The AI does the generation.
Terms You Can Safely Ignore (For Now)
Transformer architecture – How the AI brain is structured internally. You don’t need to know this any more than you need to understand combustion engineering to drive a car.
Weights and parameters – The numbers inside the AI model that determine its behavior. Measured in billions. Interesting to researchers. Irrelevant to users.
RLHF (Reinforcement Learning from Human Feedback) – How they train the AI to be helpful and safe. Good to know exists. Not something you need to understand.
Embeddings – How AI represents words as numbers internally. Important for developers. Not important for using AI as a tool.
Quantization, distillation, pruning – Ways to make AI models smaller and faster. Only matters if you’re running AI on your own hardware.
The Only Technical Concept You Actually Need
If you remember one technical thing from this article, make it this:
AI doesn’t “think.” It predicts.
When you ask AI a question, it’s not reasoning through the answer like a human. It’s predicting the most likely sequence of words based on the patterns it learned during training.
This is why it can be confidently wrong. It’s not lying. It’s predicting, and sometimes the prediction sounds right but isn’t.
This is also why context matters so much. The more context you give, the better the prediction. Because you’re narrowing down the possibilities. “Write an email” has millions of possible good responses. “Write a follow-up email to my plumber about the kitchen leak we discussed Thursday” has maybe a dozen good responses. The AI’s prediction is more accurate when the target is smaller.
Understanding this one concept will make you better at using AI than knowing every term on this list.
Your Quick Reference Card
Save this. Come back to it when you encounter a term you don’t remember.
| Term | Plain English | |——|————–| | Prompt | Your message to the AI | | Model | The AI brain (GPT-4, Claude, etc.) | | Token | Unit of text measurement (~3/4 of a word) | | Context window | How much the AI can “see” at once | | Hallucination | When AI makes something up | | LLM | The type of AI (Large Language Model) | | Fine-tuning | Specializing a general AI for specific tasks | | API | Connection between software programs | | System prompt | Standing instructions for the AI | | Temperature | Creativity dial (low = predictable, high = creative) | | RAG | AI that looks up info before answering |
That’s all you need. Everything else is detail for developers and researchers.
[Listen to “Why AI?” on our homepage] [Book Your Free Intro Session]
—
Achievementoring helps regular people build AI-powered productivity systems through 1:1 coaching, self-paced membership content, and done-for-you setup services. Because the future of personal productivity isn’t about working harder. It’s about working with intelligence.
Want help building your own AI system? Book a free intro session and see it in action. Or browse all 10 coaching sessions to see the full program.