Why this day matters
Everybody skips this day. Then they spend six months confused about why the tool keeps lying to them. A generative model is not a search engine and it is not a database. It is a prediction engine that has read a very large amount of text and images and has learned what usually comes next. That single sentence explains almost every strange thing AI will do to you for the rest of this course. Learn it today and the next fourteen days get much easier.
Where generative AI actually sits
Artificial intelligence is the whole field. Machine learning is the part that learns from data. Large language models are one kind of machine learning. Generative AI is what those models do when you ask them to produce something new.
What you learn
Generative AI
Software that produces new text, images, audio and video rather than retrieving existing ones.
Large Language Model
The engine underneath the chat box. Trained on text, it predicts the most likely next piece of language given what came before.
The prompt as a command
Your prompt is not a question. It is an instruction set. Role, task, context, constraints, format. Vague in, vague out.
Hallucination
When the model produces something fluent and confident that is simply not true. It is not lying. It has no concept of truth. It is completing a pattern.
Tokenisation
The model does not read words. It reads tokens, roughly four characters or about three quarters of an English word. Cat is one token. This is why context limits and pricing are counted the way they are.
Context window
The amount the model can hold in mind at once. Everything outside it is gone. This explains why long chats drift.
Watch first



Do the work
Open ChatGPT and ask the same question three ways. First as one sentence. Then with a role added, such as act as a Bahraini secondary school science teacher. Then with role, audience, length and format specified. Paste all three answers side by side in a document and write two lines on what changed. That document is your first artefact.
Deliberately induce a hallucination and document it. Ask for the top five sources on a niche topic in your industry, then verify every one. Screenshot what the model invented. Keep that screenshot. When a client asks whether AI can be trusted with their brand, you will show them this instead of an opinion.
A one page comparison document plus one verified hallucination example.
Exercises
Ask ChatGPT, Claude and Gemini the identical question about a topic you know deeply. Score each answer out of ten for accuracy before you read any other output, so one model cannot anchor your judgement of the next.
Find the context window limit the hard way. Paste a long article into a chat, ask three questions about the opening paragraph, then keep chatting until the model loses it. Note roughly how many turns that took.
Take one confident claim an AI gave you today and spend five minutes trying to prove it wrong with an ordinary web search. Write down whether it survived.
Brainstorm
If the model only predicts the next likely token, why does it feel like it understands you? Argue both sides.
Which tasks in your own week are safe to hand to a prediction engine, and which need a source of truth? Draw the line and defend it.
A colleague says AI lied to them. Using today's vocabulary, explain in three sentences why that framing is wrong and what actually happened.












































