✨ AI Tools & Prompting

1 What Is an AI Chatbot / LLM?

"LLM" stands for Large Language Model — a system trained on huge amounts of text to predict the most likely next word (or token) given what came before. That training gives it broad, flexible language ability, but it's fundamentally pattern-matching, not human-style understanding or reasoning — it can produce fluent, confident-sounding text that is nonetheless wrong.

The text you type to an AI tool is called a prompt — everything the model uses to generate its response comes from that prompt plus the conversation so far, not from looking anything up in real time (unless the tool explicitly adds a search/browsing feature).

2 Popular AI Tools Overview

General-purpose chat assistants like ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google) handle a wide range of conversational tasks — writing, explaining, brainstorming, analysis. Specialized tools exist too: GitHub Copilot and Cursor focus specifically on coding, integrated directly into a code editor.

Different tools can give meaningfully different answers to the same prompt, since each is built on a different underlying model trained differently — it's often worth trying more than one for an important task.

3 Free vs Paid Tiers

Most major AI tools offer a free tier with real usage limits — a capped number of messages, access to an older or smaller model, or slower response times — and a paid tier unlocking higher limits and more capable models. Exact pricing and limits change often, so check the provider's current pricing page rather than trusting older information (including this course's, over time).

For casual, everyday use, a free tier is often sufficient; heavier or professional use typically benefits from a paid plan.

4 Getting Started: Your First Conversation

Within a single conversation, an AI assistant "remembers" what you've discussed so far (up to a context limit) and uses it to inform new responses — that's why continuing an existing conversation on the same task is more effective than repeating yourself in a new one. A brand-new, separate conversation typically starts with no memory of previous separate chats, unless the tool has an explicit persistent-memory feature.

A good way to start with any new AI tool: try a simple, low-stakes question first to get a feel for its style and limitations before relying on it for something important.

5 Writing Clear, Specific Prompts

A specific prompt reliably beats a vague one. Compare these two:

❌ Vague:
Write about dogs

✅ Specific:
Write a 100-word paragraph about why dogs make good family
pets, for a pet-adoption website. Friendly, upbeat tone.

The second gives the model a real target to aim for: length, angle, audience, and tone. Specifying the desired format (a bulleted list, a table, a short paragraph) tells the model exactly how to structure its answer, instead of leaving it to guess what you actually wanted.

6 Providing Context

Background information narrows down what the model should focus on, instead of it guessing.

❌ No context:
Fix this function.

✅ With context:
This function should return the average of a list of numbers,
but it crashes on an empty list. Here's the code:

def average(nums):
    return sum(nums) / len(nums)

I'm using Python 3.11. What's the safest way to handle the
empty-list case?

For writing feedback, include the actual text plus what kind of feedback you want (tone? grammar? structure?). More context helps — but only when it's relevant; dumping in unrelated detail can dilute focus rather than sharpen it.

7 Few-Shot Examples

A few-shot prompt includes one or two examples of the input/output pattern you want before asking for a new one — showing rather than just describing the format, style, or tone. A zero-shot prompt, by contrast, gives no examples, just a direct instruction.

Convert each product name into a short, punchy tagline.

Product: Wireless earbuds → Tagline: "Cut the cord, keep the beat."
Product: Standing desk → Tagline: "Work up, not out."
Product: Reusable water bottle → Tagline: ?

The two worked examples show the model exactly the tone and length you want, so its answer for the third product is far more likely to match than if you'd just said "write a tagline."

8 Iterating on Prompts

Treat the first response as a draft, not a final answer.

You: Summarize this article in 3 bullet points.
AI: [gives 3 bullets, but they're too technical]

You: Good, but rewrite these for someone with no background
in the topic — avoid jargon.
AI: [gives a simpler version]

You: Perfect. Now make bullet 2 one sentence shorter.

Each follow-up refines the previous answer rather than starting over — for complex tasks, a few rounds of refinement is usually more effective than trying to write one perfect prompt upfront. You can even ask the AI to critique or find flaws in its own previous answer, surfacing issues you can then ask it to fix.

9 AI for Writing & Editing

AI is genuinely useful for drafting, rewriting for tone or clarity, grammar checking, and brainstorming — but it can't be relied on to verify facts with complete certainty. Use it to produce a starting draft, then read it over to make sure it actually reflects what you mean and sounds like you before sending or publishing it.

Brainstorming is one of the tasks AI is generally best suited for — generating a wide spread of angles or ideas quickly, for you to filter and build on.

10 AI for Research & Summarization

Every model has a knowledge cutoff — a point in time after which it has no training data or awareness of events — and can generate confident-sounding but incorrect information (a "hallucination," covered in Module 5). For summarizing a document, provide the actual text rather than a vague description of it, so the summary is grounded in the real source.

If an AI gives you a specific statistic, date, or citation, independently verify it against the original source before relying on it — this matters most for anything important.

11 AI for Learning & Studying

AI can act like a patient tutor — explaining a concept step by step at your pace, and answering follow-up questions until it clicks. Asking it to quiz you rather than just re-explain material supports active recall, the same principle behind the practice quizzes across this site's Learn Hub and exam-prep courses.

Explaining a concept back to an AI (or having it check your own explanation) reveals gaps in your understanding, similar to teaching it to someone else. Still, cross-check its explanations against your actual course material, especially for anything exam-relevant.

12 AI for Planning & Organization

AI can help break a large goal into smaller, ordered steps, or draft an initial to-do list or schedule for you to refine. It's a good starting point, not a guarantee — an AI-generated schedule doesn't automatically fit your actual constraints, so you still need to review and adjust it.

Final decisions on priorities and deadlines should ultimately be made by you, using AI's suggestions as input rather than as the final authority.

13 AI Coding Assistants Overview

Tools like GitHub Copilot and Cursor suggest or generate code directly within your editor, based on the surrounding context. Agentic coding assistants like Claude Code go further — they can perform multi-step tasks: reading files across a project, running commands, and making edits, not just suggesting a single line at a time.

None of these guarantee bug-free code — every suggestion still needs to be reviewed, tested, and understood before you rely on it, which is the subject of the next two topics.

14 Writing Effective Coding Prompts

❌ Vague:
Write a function to validate emails.

✅ Specific:
Write a TypeScript function `isValidEmail(input: string): boolean`
that checks for a valid email format. It should return false for
empty strings, strings with no "@", and strings with spaces.
Add 3 example calls showing true/false results as comments.

Specify the programming language/framework and the desired behavior, including edge cases, rather than leaving it to the model to guess. Pasting relevant existing code into your prompt gives the AI real context to match — your project's actual conventions — rather than generating in a vacuum. Asking the AI to explain what it's doing as it writes helps you actually understand and verify the code, instead of blindly accepting whatever it produces.

15 Debugging with AI

❌ Vague:
My code doesn't work, please fix it.

✅ Specific:
I'm getting "TypeError: Cannot read properties of undefined
(reading 'map')" on line 12. I expected `users` to be an array
here, but it seems to be undefined on first render. Relevant code:

function UserList({ users }) {
  return users.map(u => 
{u.name}
);
}

What's the safest way to handle `users` being undefined initially?

The single most useful thing you can give an AI for debugging help is the exact error message plus the relevant code — not a vague description of "it's not working." Describing expected behavior versus actual behavior clarifies precisely what's wrong, not just that something is. Never trust a suggested fix without testing it — run it and confirm it actually resolves the issue.

16 Reviewing AI-Generated Code Critically

Merging AI-generated code without reading it is risky — it might contain bugs, security issues, or reference an API that doesn't actually exist (a code "hallucination" — the model confidently invents a plausible-looking function or method that was never real). Before trusting a claim about how a library function works, check the library's actual documentation.

Always run your tests after applying AI-suggested changes — that verifies the change actually works as intended, rather than just looking plausible on the page.

17 Hallucinations & Fact-Checking

An AI hallucination is confidently stated information that is actually incorrect or entirely made up — a citation that doesn't exist, a statistic that was never real, a function that isn't part of the library. This happens because the model generates plausible-sounding text from learned patterns, with no built-in fact-checking mechanism.

The habit that protects you: whenever an AI cites a specific fact, statistic, date, or source, look up the original yourself before relying on it — especially for anything that matters.

18 Privacy & Data Considerations

Before pasting sensitive personal, confidential, or proprietary information into an AI tool, think about whether that data should be shared with a third-party service at all — check the provider's data-use policy. By default, some tools may use submitted conversations to help improve future models, depending on the provider and your account settings.

Many providers offer a setting to opt out of having your conversations used for training — worth checking if you're a regular user. Before using an AI tool with work or confidential information, check both your organization's policy and the AI provider's terms.

19 Bias & Ethical Use

AI output can reflect biases present in the huge datasets of human-generated text it was trained on. For consequential decisions — hiring, evaluations, anything affecting other people — AI output should be treated as one input requiring human judgment and oversight, not the final decision-maker.

Using AI doesn't remove your responsibility for the final output — whoever uses, publishes, or acts on AI-generated content remains responsible for it, the same as with any other tool.

20 When Not to Use AI

For high-stakes situations — medical, legal, or financial decisions — treat AI as a starting point for questions to bring to a qualified professional, not a replacement for one. Relying solely on AI for a final authoritative answer on something critical is risky, since it can be confidently wrong with no reliable way to gauge its own certainty.

A sensible general principle: use AI tools to assist and speed up your own work and judgment, not to replace it — and never publish AI-generated content for a real audience without reviewing it yourself first.