Day 2: The Eight Words That Unlock Everything

The Concept

There is a particular kind of frustration that comes from trying to learn something when every other sentence contains a word you do not quite understand. AI writing is full of that experience. People talk about prompting, hallucinations, context windows, and large language models as though the meaning is obvious — and if you have not been immersed in the field, the whole thing can feel like a different language.

Today solves that. You do not need to understand the engineering behind AI to use it effectively, any more than you need to understand how a combustion engine works to drive a car. But eight words will unlock almost every tutorial, article, product announcement, and conversation about AI you will encounter from here on.

The eight terms

LLM — Large Language Model. The technical name for the type of AI you are using when you talk to ChatGPT, Claude, or Gemini. Large refers to the scale of its training. Language is what it was trained to produce. Model is the underlying system. When people say "the model," this is what they mean.

Prompt. The text you type when you interact with AI. A prompt is your instruction. "Summarise this article" is a prompt. "Help me write a difficult email to my landlord" is a prompt. Getting better at writing prompts is one of the most learnable and high-value skills in this course — Day 5 is dedicated to it entirely.

Hallucination. When AI generates something false with the same confident fluency it uses for something true. It might invent a statistic, create a plausible-sounding but non-existent book, or describe an event that never happened. Understanding this term early is important because the risk is not obvious — wrong AI output does not look different from correct AI output.

Context window. The total amount of text AI can work with in a single conversation. Your messages, AI's responses, any documents you have shared — all of this counts toward the limit. When a conversation becomes very long or a document is very large, you may need to break it up. Most modern tools have generous limits, but it is worth knowing the concept exists.

Token. The unit AI uses to process text — roughly three or four characters, or about three-quarters of a word. Token limits determine how much AI can read or generate in one go. You will rarely hit limits in everyday use, but this word appears frequently in product descriptions and pricing.

Training data. The material the AI was trained on — books, articles, websites, code, and more. The composition of training data shapes everything: what the model knows, what gaps it has, and what assumptions and biases it reflects. Models trained predominantly on English-language internet text will reflect the knowledge, perspective, and blind spots of that source material.

Model. The specific version of an AI system. GPT-4o, Claude Sonnet, Gemini 1.5 Flash — these are all different models with different capabilities and costs. Within the same company, there is usually a faster lighter version and a more powerful one. Knowing which model you are using helps you understand what to expect.

Generative AI. AI that generates new content — text, images, code, audio, video — rather than simply classifying or retrieving existing information. This distinguishes the tools in this course from older AI systems like spam filters or recommendation algorithms that you have been using for years without thinking of them as AI.

Making the vocabulary stick

These eight terms cover the vast majority of what you will encounter. You do not need to memorise them — you need to recognise them when they appear and know roughly what they mean. After today, you will. The most effective way to move vocabulary from recognition to real understanding is to use it immediately in your own words, which is exactly what today's task asks you to do. The two terms you write out yourself will stay with you far longer than any glossary you read and forget.