Introduction to
Generative AI

Generative AI is the most talked-about technology of our lifetime, and most of what is said about it is either breathless hype or quiet fear. Neither helps you. This course is the middle path. It will teach you, plainly and honestly, what this technology is, what it can and cannot do, and how to make it genuinely useful in your work and your life without being fooled by it.

You do not need any technical background. If you can send a message and think carefully, you can master this. Every idea comes with something to try, because you understand AI far better after you have poked it than after you have read about it. At the end there is a thirty-question exam, and passing it with eighty percent earns you a certificate signed on behalf of the Council.

Let me demystify the whole thing in one idea. A large language model, the engine inside ChatGPT, Claude, Gemini and the rest, does one deceptively simple thing. It predicts the next piece of text. You give it some words, and it asks, based on everything I have read, what word most likely comes next. Then it does that again, and again, one small chunk at a time, until it has written a whole answer. Those small chunks are called tokens, roughly a word or part of a word each.

That is genuinely most of it. It is not looking up answers in a database of facts. It is not thinking the way you think. It is an extraordinarily well-read prediction machine. Play the model yourself below and feel how it works.

This one idea explains almost everything that follows. Because the model predicts likely text rather than retrieving verified facts, it is brilliant at fluent writing and occasionally confident about things that are simply not true. Hold onto that. It is the key to using it wisely.

The single skill that separates people who get gold from AI and people who get mush is the prompt. A prompt is just the instruction you give. Vague in, vague out. The good news is that a strong prompt follows a simple recipe: give it context, a clear task, the audience, and the format or length you want. You do not need clever tricks. You need to be specific in the way you would be with a talented new intern who knows a lot but cannot read your mind.

Watch how the instruction sharpens as you add each piece. That final prompt will produce something you can almost use as-is, while the bare version produces something you will have to fight with. This is the whole game. When an AI answer disappoints you, the fastest fix is almost always a better prompt, not a better model.

Many AI tools let you turn a dial called temperature. It controls how adventurous the model is. Turn it low and the model plays it safe, giving focused, predictable, repeatable answers. Turn it high and it takes more chances, giving varied, surprising, creative output. Neither is better. The right setting depends on the job. Drag the dial below and watch the same request change character.

The lesson is control. When you need precision, such as pulling exact figures from a report, keep it low. When you want a hundred ideas for a campaign, turn it up. Knowing which to reach for is the mark of someone who actually understands the tool.

Generative AI is not good at everything, and pretending otherwise is how people get hurt. The clearest way to think about it is by stakes. It shines on low-stakes creative and drafting work, where a mistake costs nothing and a human polishes the result anyway. It is useful but needs a check on medium-stakes work. And it should never be the final word on high-stakes decisions that a person will act on, such as medical, legal, or financial advice. Sort these tasks by how much you can trust the AI to work alone.

We met hallucination in module one. Now let us make you good at catching it. Remember why it happens: the model predicts plausible text, and plausible is not the same as true. It will invent a statistic, a law, a quote, or a source, and deliver it with total confidence, because it has no sense of doubt. It is not lying. It genuinely cannot tell the difference. That is your job now.

So how do you defend yourself? Two habits. First, grounding: when you need facts, give the model the real source material and tell it to answer only from what you provided, then still check. Second, verification: anything you will act on, a figure, a citation, a legal point, gets confirmed against a trusted source before you use it. Asking the model are you sure does not count. It can be confidently wrong twice.

Power without care causes harm, and generative AI is genuinely powerful. Responsible use is not a set of rules imposed from above. It is a handful of habits that protect you, your organisation, and the people you serve. Three matter most. Protect private data and never paste personal or confidential information into public tools. Stay transparent and take human responsibility for anything you publish with AI's help. And watch for bias, because a model trained on human writing can quietly reproduce human prejudice.

The same technology that writes your marketing copy can fake a human being. A deepfake is synthetic media, a video or a voice, generated to convincingly imitate a real person. Combined with cheap voice cloning, this has already changed the fraud landscape. A scammer no longer needs to guess your boss's writing style. They can clone their voice from a few seconds of audio and call your accounts department with an urgent request to wire money. Your defence is a simple, unbreakable rule.

Beyond fraud there is a subtler technical risk worth naming: prompt injection. Because AI systems read and act on text, an attacker can hide malicious instructions inside a web page or document, and a naive AI assistant may obey them. You do not need to defend against this yourself yet, but knowing the term means you will recognise the danger when you start connecting AI to real systems.

Everything so far leads to this. Generative AI is not something happening to the Caribbean from the outside. It is a tool we can pick up today, and it favours exactly what we have: creativity, small nimble teams, and a hunger to do more with less. The opportunity is real and it is now. Here is where it is already paying off across the region.

A pattern runs through every one of those. AI does the fast first draft, and a Caribbean human brings the judgement, the culture, and the final word. That is not a limitation. That is the healthiest way to work with this technology anywhere in the world, and it happens to play to our strengths.

Thirty questions from across the whole course. Score eighty percent or higher to earn your certificate. Retake it as often as you like, and read the explanation on anything you miss, because a wrong answer is just one more thing learned.