AI for Everyone — A Practical Guide

Lesson 01

What Is AI — and What It Isn't

By the end of this lesson, you'll be able to describe AI in plain terms and stop believing the science-fiction version.

Artificial Intelligence is software that has been trained on enormous amounts of text, images, or other data — and has learned to recognize patterns in that data. When you ask it a question, it uses those patterns to generate a response. It is not thinking. It is not conscious. It is doing something more like very sophisticated pattern-matching at a scale that can feel uncanny.

The AI tools you'll use day-to-day — ChatGPT, Claude, Gemini — are a specific kind called Large Language Models (LLMs). They were trained on a huge portion of the internet, books, and other text. Their job is to predict what words should come next, given the words you gave them. That single trick, applied at enormous scale, produces something that feels surprisingly intelligent.

What AI is not: it doesn't have goals, feelings, or intentions. It doesn't remember you between conversations (unless told to). It can be wrong — confidently, fluently wrong — and it won't always tell you so.

Try This — Exercise 1

Open Claude or ChatGPT (free — signing up takes about two minutes with any email address). Once you're in, type exactly this:

"Explain what you are in three sentences, as if I'm twelve years old."

Read the answer carefully. Does it claim to think, feel, or know things? That's your starting point for understanding what these tools believe about themselves — and where they're stretching the truth.

Exercise 2 — The honesty test

In the same chat window, ask:

"Do you actually understand what I'm saying, or are you predicting words? Be honest about the difference."

Notice how it handles this. Different tools handle it differently — some are more candid, some deflect. This tells you a lot about how each tool has been designed to present itself. There is no right or wrong answer to watch for; the interesting part is the variation.

Exercise 3 — The wrong answer test

Ask an AI something you already know well — your profession, your hometown, a hobby. See whether it gets things right, fills gaps with plausible-sounding fiction, or admits it doesn't know. This exercise builds your instinct for when to trust AI output and when to check it.

"What do you know about [your city or town]? Give me five facts about it."

Verify each fact. You'll likely catch at least one thing that sounds convincing but is inaccurate or invented. This is not a flaw unique to one tool — all of them do it.

How AI was trained

Before an AI tool can answer questions, it goes through a process called training. Imagine reading billions of documents — books, websites, articles, forum posts — and slowly learning the patterns of how language works. That's roughly what happens, except machines do it in weeks using thousands of processors running simultaneously.

During training, the model makes predictions ("given these words, what comes next?") and is corrected millions of times until its predictions become reliably good. It never memorizes the internet word-for-word — it learns the patterns behind language. That's why it can write a poem in the style of Shakespeare even if no prompt ever asked for exactly that.

After that basic training, most tools go through a second phase called RLHF — Reinforcement Learning from Human Feedback. Human raters review the model's responses and flag which ones are better, more helpful, or safer. The model learns to prefer those. That's why ChatGPT, Claude, and Gemini feel polite and well-organized rather than like raw, unpredictable text machines.

Why it doesn't always know what it doesn't know

Because the model was trained to produce the most plausible next word, it doesn't have a built-in uncertainty signal. It generates a confident-sounding response even when it is reaching into territory it has no real knowledge of. This is called hallucination — the model fills gaps with something that sounds right. Always verify anything that matters.

💡

AI is pattern-matching software trained on data. It generates plausible responses — not necessarily true ones. The moment you remember that, you'll use it much more wisely.

Lesson 02

What AI Can and Can't Do

By the end of this lesson, you'll know where AI genuinely helps, where it falls flat, and why — so you stop being surprised by either.

AI chat tools are remarkably good at tasks that involve working with language: summarizing, drafting, explaining, translating, brainstorming, and editing. They're also useful for structured thinking — helping you outline a plan, organize your thoughts, or see a problem from a new angle.

But they fail in predictable ways. They have a knowledge cutoff — they don't know about events after their training ended (typically one to two years ago). They have no memory between sessions — every conversation starts completely fresh. And they hallucinate: they sometimes produce facts, names, dates, or citations that sound authoritative but are simply made up. That last one is the most important limitation to internalize.

Where AI shines

Drafting and editing text — emails, messages, documents, cover letters
Explaining complicated topics in plain, accessible language
Brainstorming ideas when you feel stuck or need a fresh perspective
Translating between languages quickly and accurately
Summarizing long documents, articles, or notes into key points
Writing, fixing, or explaining code
Answering well-established general knowledge questions

Where it fails

Precise facts, statistics, or current events — always verify these independently
Legal, medical, or financial advice — it can inform, it cannot advise
Tasks requiring real-time information (unless the tool has a live search feature)
Anything requiring true judgment, accountability, or genuine common sense
Remembering what you said in a previous conversation

The most important thing to remember about AI

AI will not say "I don't know" when it should. It produces a fluent, confident-sounding answer regardless of whether it actually has one. Never trust AI-generated facts in anything consequential without verifying from an independent source.

Try This — Exercise 1 (hallucination in action)

Open any AI chat tool. Type exactly this:

"Can you recommend three academic papers about the effects of social media on sleep in teenagers? Include the authors, journal names, and publication years."

Then take one of the papers it suggests and search for it on Google Scholar (scholar.google.com) or your library's search tool. Odds are good that at least one citation is partially or entirely invented — a real-sounding title, plausible authors, but the paper doesn't exist. Experiencing this firsthand is more useful than being told about it.

Exercise 2 — Test the knowledge cutoff

Ask an AI tool about something you know happened recently — a news story, a product launch, a sports result from the last year. Notice whether it knows about it, admits it doesn't, or confidently gives you stale or wrong information. This tells you exactly how current its knowledge actually is.

"What happened in the most recent [a major news event you follow]? Give me the key details."

Exercise 2 — The memory gap, made concrete

Start a new chat and tell the AI something specific about yourself — your name, your job, a preference. Have a short exchange. Then close that chat completely and open a brand new one. Ask the AI to recall what you told it. It will have no idea. The slate is completely clean. Some tools (Claude and ChatGPT with memory enabled) can be configured to remember things across sessions — but that's an opt-in setting, not the default behavior.

Why hallucination happens

When you ask a language model a question, it doesn't look things up — it generates an answer word by word, choosing the most statistically plausible continuation of the text. If the true answer isn't strongly represented in its training data, it doesn't leave a blank — it fills the gap with something that sounds right.

Think of it like a confident student who studied hard but doesn't know the answer to a specific exam question. Instead of writing "I don't know," they write something plausible based on what they do know. The model has no shame mechanism — it just keeps generating.

Hallucination is worst with: specific names and dates, citations and sources, statistics, local or niche information, and anything recent. It's less problematic for general explanations, creative tasks, and well-established concepts where the training data is dense and consistent.

The context window — why AI forgets mid-conversation

Every AI conversation has a limited "working memory" called a context window — the amount of text the model can hold in mind at once. In older or cheaper models, this is relatively small. Modern models can handle very large amounts. But it is always finite, and once a conversation grows long enough, earlier parts start to fall out. This is why AI can seem to forget something you said ten exchanges ago. It's not selective — it literally no longer has access to that part of the conversation.

💡

AI is a powerful first-draft tool, not a final authority. Use it to get started, generate ideas, and process language — then apply your own judgment before trusting the result.

Lesson 03

Free Tools Worth Knowing

By the end of this lesson, you'll know which tool to reach for, what each one does best, and exactly where to go to start using it.

All the tools below have free tiers that are good enough to get real work done. You will need to create a free account for each one — it typically takes two minutes with an email address, and you won't need a credit card for any of the free tiers listed here.

Chat & Writing

ChatGPT

Free

Made by OpenAI. The most widely known AI assistant and the one most people try first. The free tier uses GPT-4o mini, which handles most everyday tasks well. The paid plan ($20/month) unlocks significantly more power for complex tasks.

Best for: All-purpose everyday tasks — writing emails, answering questions, brainstorming, summarizing. The most polished general-purpose experience with the largest base of tutorials and guides online.

→ Open ChatGPT (chat.openai.com)

First time? Click "Sign up" — works with Google, Microsoft, or any email address.

Claude

Free

Made by Anthropic. Widely regarded as the strongest AI for long, nuanced writing tasks — it handles long documents, careful reasoning, and subtle tone better than most competitors. The free tier is genuinely capable, not a limited demo.

Best for: Detailed writing and editing, reading and summarizing long documents, reasoning through complex topics, tasks where tone, care, and precision matter most. Also strong for sensitive or difficult conversations.

→ Open Claude (claude.ai)

First time? Click "Sign up" — works with Google account or any email address.

Gemini

Free

Google's AI assistant. Because it connects to Google Search, it can access current information and recent events in a way that tools relying solely on training data cannot. Also integrates directly with Gmail, Google Docs, and Google Drive if you use those.

Best for: Questions about current events, recent news, and anything requiring up-to-date information. Also useful if you're already inside the Google ecosystem and want AI help without leaving your existing tools.

→ Open Gemini (gemini.google.com)

First time? Requires a Google account — any Gmail address works.

Perplexity

Free

A search engine with an AI layer on top. Unlike other tools, every answer comes with cited, clickable sources — you're never left trusting an invisible output. It searches the web in real time for each question, then synthesizes the results into a clear answer.

Best for: Research questions, fact-checking, and any situation where you want to verify where information came from. The most transparent of the major AI tools — you can always trace back to the original source.

→ Open Perplexity (perplexity.ai)

First time? You can use it immediately without an account. Signing up saves your search history.

Image generation

These tools generate images from a text description. You describe what you want in plain words and the tool creates it from scratch. All free options below require a free account.

DALL·E via ChatGPT

Free

Available directly inside ChatGPT — no separate account or setup needed. Just describe the image you want in plain English and it generates it. Free users get a limited number of generations per day. Results are clean, realistic, and easy to get started with.

Best for: Beginners who want their first image generation experience without any setup. Quick illustrations, concept visuals, and images to accompany writing. The easiest entry point into AI image tools.

→ Open ChatGPT and ask it to make an image

Uses your existing ChatGPT account — no extra signup needed.

Adobe Firefly

Free

Adobe's image generator, trained exclusively on licensed and public-domain content — not scraped from artists without permission. This makes it the safest choice if you plan to use generated images in published work, professional documents, or anything public-facing. Free users receive 25 credits per month.

Best for: Professionally safe image generation. If the images you create might end up in a presentation, on a website, in marketing materials, or anywhere public, Firefly is the responsible default.

→ Open Adobe Firefly (firefly.adobe.com)

First time? Free Adobe account — sign up with any email address, no Creative Cloud subscription needed.

Flow by Google

Free

Google's AI creative tool powered by their Imagen and Veo models. Generates high-quality images from text descriptions and can also produce short video clips — both from the same platform. Clean, straightforward interface with no complicated settings to navigate.

Best for: High-quality photorealistic images and short AI video clips. The only free tool that combines image and video generation in one place, making it particularly useful if you want to experiment with motion as well as stills.

→ Open Flow by Google (flow.google)

Requires a Google account. Some features may have a waitlist — sign in to check your access.

Midjourney

Paid only

The most artistically powerful image generator available. Produces stunning, stylized results that feel genuinely designed rather than generated. Paid only ($10/month minimum) and operates through Discord, which adds a small learning curve. Not a starting point — a graduation point.

Best for: Creative, artistic, and cinematic imagery where quality matters most. The tool professional designers, illustrators, and creative directors reach for when the free tools aren't enough. Worth exploring once you're comfortable with the basics.

→ Open Midjourney (midjourney.com)

Requires a Discord account and a paid Midjourney subscription. Start with the free tools above first.

Try This — Exercise 1

Pick any two chat tools from the list above. If you don't have accounts yet, sign up for both now — it takes about two minutes each. Then ask them the exact same question:

"I need to write a short message to a neighbour politely asking them to keep the noise down at night. Can you help me draft it?"

Compare the two answers side by side. Which felt more natural? Which was too formal or too casual? Which would you actually send? This exercise builds a feel for each tool's default personality — which will help you know which one to reach for in future.

Exercise 2 — Use Perplexity for a real research question

Think of something you've been meaning to look up — a health question, a product comparison, a how-to for something around the house. Go to perplexity.ai and ask it. When the answer comes back, click on at least one of the source links it provides. Check that the source actually says what Perplexity claims it says. This is the habit that separates careful AI users from careless ones.

Exercise 3 — Generate your first image

Open ChatGPT or Adobe Firefly. Write a description of something specific you'd like to see:

"A cosy bookshop at night, warm lamplight glowing in the window, light rain on the street outside, painted in a loose watercolour style."

Look carefully at what comes back. Notice what the tool gets right and what it distorts or misses. Then change one word in your description — "watercolour" to "oil painting," or "night" to "golden hour" — and generate again. Image prompting rewards specificity: the more detail you give, the more intentional the result.

Exercise 4 — Test Gemini's search advantage

Ask Gemini and ChatGPT the exact same question about something that happened in the last few months. A news story you followed, a sports result, a product announcement. Notice which one gives you current, accurate information and which one either gets it wrong or admits it doesn't know. This makes the live-search advantage concrete rather than theoretical.

Why different tools feel different

ChatGPT, Claude, and Gemini are all large language models, but built by different companies with different priorities. OpenAI (ChatGPT) focused early on broad capability and consumer accessibility. Anthropic (Claude) focused heavily on safety and careful reasoning — their model tends to be more thoughtful and is less likely to produce harmful content, though it can occasionally be more cautious than necessary. Google (Gemini) integrated search from the start, giving it a real-time information advantage but a slightly different personality — more retrieval-focused than generative.

The underlying models also differ in how they were fine-tuned through human feedback — the second training phase that shapes personality. That's why the same question asked to three different tools produces three different tones, lengths, and styles. None of them is universally "better" — they're better at different things.

Free vs. paid — what do you actually lose?

The free tier of most tools uses a smaller or older version of the model. For everyday tasks — writing help, quick questions, brainstorming — this is usually fine and you won't notice the difference. Where the gap shows: complex multi-step reasoning, handling very long documents, creative tasks requiring real subtlety, and anything where accuracy matters most. If you find yourself hitting frustrating limits on the free tier, a single paid subscription ($20/month for ChatGPT or Claude) covers most people's needs. You don't need all of them paid — pick the one you use most.

💡

You don't need to pick one tool. Try a couple, find which response style suits you, and use different ones for different tasks. All of the free tiers are genuinely usable — start experimenting.

Lesson 04

How to Prompt Better

By the end of this lesson, you'll know how to get dramatically better results from any AI tool just by changing how you ask.

The single biggest factor in the quality of what AI produces is the quality of what you give it. A vague question gets a vague answer. A specific, context-rich question gets something genuinely useful. This is called prompting — and it's the most valuable practical skill in AI use.

You don't need a course or a certification to get good at this. You just need to understand what information the AI is missing and give it that.

The four ingredients of a good prompt

Context: Who are you, what's the situation, what does the AI need to know to help you?
Task: What exactly do you want it to do? Be as specific as possible.
Format: How should the answer look? A list? Short paragraphs? Formal or casual?
Constraints: What should it avoid? What tone? What should it not assume?

See the difference

Weak prompt

"Write me an email."

Strong prompt

"Write a short, friendly email to a colleague named Sarah letting her know I'll be 20 minutes late to our 3pm meeting today. Keep it under five sentences. Don't over-apologize. End with a question asking whether she'd prefer to push the meeting back or keep the time."

Same request — completely different result. The second prompt gives the AI a person, a situation, a tone, a length limit, something to avoid, and a specific ending. There is almost nothing left to guess.

Useful techniques

Give it a role: "Act as a patient teacher explaining this to a complete beginner."
Ask it to try again: "That's too formal. Make it more conversational and cut it in half."
Think out loud with it: Paste in your rough notes and say "help me turn this into a proper paragraph."
Ask for options: "Give me three different ways to phrase the opening sentence."
Tell it what not to do: "Don't use bullet points. Don't start with the word 'I'. Don't be sycophantic."
Set the audience: "Explain this as if I'm not a technical person" — or the reverse.

Try This — Exercise 1

Do this in two rounds. Open Claude or ChatGPT.

Round 1 — weak prompt: Type only: "Help me write something for work." See what the AI does with nothing to go on. It will either ask you a clarifying question or produce something completely generic.
Round 2 — strong prompt: Now describe a real work task you actually have coming up. Use the four ingredients: who you are, what you need, the format, and at least one constraint about tone or length.
Compare the two results. The gap between them is entirely explained by what you gave the AI to work with.

Exercise 2 — Practice redirecting

Ask an AI to write a short bio about you (give it a few real facts). When the response comes back, tell it exactly what's wrong:

"That's too formal. I want it to sound warmer and more personal, like something a friend might say introducing me. Also remove the part about [something it included that you didn't want]."

Keep redirecting until it gets close. Notice how much better the results get with each round. This back-and-forth is the normal workflow — the first response is almost never the final one, and that's fine.

Exercise 3 — Roleplay for preparation

Pick a situation you find uncomfortable — a difficult conversation at work, a negotiation, explaining a decision to someone who won't like it. Ask the AI to roleplay as the other person:

"I need to tell my manager I'm resigning. Play the role of a manager who's surprised and a bit frustrated. Let me practice what I'd say, and respond the way a real manager might. Don't let me off easy."

This is one of the most underused AI applications: low-stakes rehearsal for high-stakes conversations. You can try different approaches and see what lands without any real consequences.

Exercise 4 — The audience test

Ask the same question three times, each time with a different audience instruction. For example, explaining your job:

"Explain what a [your job title] does to a 10-year-old."
"Explain what a [your job title] does to a recent graduate considering the field."
"Explain what a [your job title] does to a senior executive who needs a one-paragraph summary."

Compare all three answers. This exercise makes visible how much of the AI's output is shaped by who it thinks it's talking to — and how much control you actually have over that.

Chain-of-thought prompting

One of the most reliable ways to improve AI reasoning is to ask it to think step by step before answering. Simply adding "Think through this step by step before giving me your answer" to any prompt involving reasoning, math, or multi-part decisions can significantly improve accuracy. The model "shows its work" and tends to catch its own errors in the process.

Few-shot prompting

If you want the AI to produce output in a very specific style or format, the most reliable method is to show it examples before asking. For instance: "Here are three email subject lines I like: [example 1], [example 2], [example 3]. Now write five more in exactly the same style for a newsletter about [topic]." Giving examples rather than describing what you want in the abstract almost always produces better-matched results.

Persistent instructions (system prompts)

Most chat tools let you set instructions that apply to every message in a conversation — or even across all conversations. For example: "Always keep your answers under 150 words. Never use bullet points. Write in a warm but professional tone." In ChatGPT this is called "Custom Instructions" (Settings → Custom Instructions). In Claude you can set it in conversation settings. This saves you from repeating the same setup at the start of every chat.

💡

Prompting is a conversation, not a command. Give the AI context, be specific about what you want, and redirect it when the first answer misses the mark. The skill builds quickly with practice.

Lesson 05

Practical Uses in Real Life

By the end of this lesson, you'll have a concrete list of ways AI can save you time or effort starting today — and you'll have actually tried one.

The best way to build the habit is to find one use case where AI saves you real time and start there. The lists below are organized by context. Don't try all of them — pick two or three that apply to your actual week right now.

At work

Drafting emails you've been putting off — give it bullet points, get a full draft back
Summarizing long reports or meeting notes into key points in under a minute
Preparing for a difficult conversation by asking it to help you organize your argument
Creating a first draft of any document: job posting, proposal, policy, announcement
Translating materials for colleagues or clients who speak other languages
Explaining a technical concept in plain language for a non-technical audience

At home

Meal planning — tell it what's in your fridge, ask for three dinner ideas and a shopping list
Explaining concepts to kids (not doing their homework — explaining what they don't understand)
Planning a trip — give it your dates, interests, and budget; ask for a day-by-day itinerary
Drafting personal letters, thank-you notes, or difficult messages when words don't come easily
Researching a purchase — ask for an honest comparison of two or three products you're considering

For learning

Ask it to explain anything you don't understand — at exactly the level you need, in your words
Have it quiz you on a topic you're studying
Use it as a thinking partner to work through a decision or problem you're stuck on
Get feedback on your writing, a presentation, or a cover letter

Try This — Exercise 1 (do this now, with something real)

Open Claude or ChatGPT. Pick one real task from the lists above that you actually have to do this week — not a test task, a real one. Then:

Describe the task with full context: who's involved, what the situation is, what outcome you want.
Read what comes back. Don't accept it as final — read it critically, as you would a first draft from a colleague.
Tell it one thing to change. ("Make it shorter." / "Less formal." / "Add a question at the end.")
Use the revised version as your starting point for the real thing.

Notice how long the whole process took. That's your comparison point — not "did AI do this perfectly?" but "did this save me time and friction?"

Exercise 2 — Summarize something long you've been avoiding

Find a long email chain, article, or document you haven't gotten around to reading. Copy the full text. Open Claude (it handles long documents especially well) and paste it in with this prompt:

"Summarize this in five bullet points. For each point, tell me whether there's anything I need to act on or respond to."

This is one of the most immediately useful AI applications. Paste long content, ask for a structured summary — you'll use this habit every week once you start.

Exercise 3 — Plan something real with real constraints

Ask AI to help you plan something you've been meaning to organize — a weekend trip, a dinner party, a home improvement project. Give it actual constraints:

"Help me plan a two-day road trip for two adults and one child (age 9) from [your city]. Budget around $300 total. The child likes nature and animals. We don't want to drive more than 2 hours each way. Give me a specific itinerary with suggestions for where to stay and two or three activities per day."

The more specific your constraints, the more useful the output. Vague trips get vague itineraries. Real constraints get real plans.

Exercise 4 — Use it as a thinking partner

Describe a decision you're genuinely trying to make. Ask the AI not to tell you what to do, but to help you think it through:

"I'm trying to decide whether to [situation]. Here are the factors I'm weighing: [list them honestly]. I'm not looking for a recommendation. I want you to ask me questions that help me clarify what I actually want."

This surprises most people. AI as a structured thinking partner — rather than just an answer machine — is one of its most valuable and least obvious applications. It asks questions you weren't asking yourself.

The jobs question

It's hard to discuss AI practical uses without someone asking: "Is it going to take my job?" The honest answer is nuanced and less dramatic than headlines suggest. AI is very good at the mechanical parts of knowledge work — drafting, formatting, summarizing, translating. It is not good at judgment, relationships, accountability, context-specific expertise, or anything requiring deep knowledge of a particular situation. The roles most affected are those where the mechanical parts make up the majority of the work. The roles least affected are those where judgment dominates.

What's more likely for most people than replacement: a gradual shift where people who use AI well produce more output with less effort — and those who don't find themselves working alongside, and competing with, people who do. The skill gap between "can use AI effectively" and "can't" is growing fast. This course is about closing that gap.

What companies are actually doing with AI right now

Most organizations are not replacing people wholesale. They're finding that AI handles first drafts, routine summaries, customer service templates, and data analysis faster and cheaper — which means fewer people are needed for those specific tasks, but the people who remain need to be good at directing the AI, checking its output, and applying judgment to what it produces. The skill most in demand isn't knowing how to build AI — it's knowing how to use it efficiently and critically in your specific work context.

💡

Start with one use case that solves a real problem you have this week. The habit builds itself once you feel the time saved. Don't try to use AI for everything at once.

Lesson 06

Risks & Safety

By the end of this lesson, you'll know the real risks to watch for — not the science-fiction ones — and have a simple set of rules to stay on the right side of them.

Most of what you've heard about AI risks involves robots, sentient machines, or job apocalypses. Those make for good movies. The actual risks for everyday users are quieter and more practical — but genuinely worth understanding and taking seriously.

The real risks

Hallucination — AI making things up

AI can produce confident, fluent, completely false information: fake statistics, invented citations, wrong dates, people who don't exist. This isn't a bug waiting to be fixed — it's a structural feature of how these models work. Never use AI-generated facts in anything important without verifying them from an independent source.

Privacy — what happens to what you type

Anything you type into a public AI chat tool may be stored, reviewed by employees, or used to improve the model. Do not paste confidential work documents, personal data, passwords, financial details, medical information, or anything you wouldn't want a stranger to read. Most tools have privacy settings — take five minutes to find and review them. Enterprise versions of these tools have stronger data protections if your company uses them.

Over-reliance — trusting without verifying

It's easy to start accepting AI outputs without reading them critically. This is particularly risky for medical, legal, or financial questions — areas where AI can give you a plausible-sounding answer that points you in the wrong direction. AI can inform and summarize; it cannot replace a professional for decisions that matter.

Bias and blind spots

AI models reflect the data they were trained on — which reflects human biases, cultural gaps, and Western-centric perspectives. If you're using AI to draft something that goes out under your name, read it critically before sending. It is a tool that amplifies your voice, not a neutral third party.

Deepfakes and AI-generated misinformation

Image, audio, and video tools can now create convincing fake media: a photo of something that never happened, a voice recording of someone saying something they never said. Be more sceptical of dramatic images or videos online — especially around elections, breaking news, or anything that triggers a strong emotional reaction. If it seems designed to outrage, verify before sharing.

Simple rules to live by

Never paste confidential, personal, or sensitive information into a public AI tool
Verify important facts from a second source before using them in anything consequential
Treat AI output as a first draft, not a final answer — especially for professional or public use
For health, legal, or financial decisions: consult an actual professional
Be sceptical of dramatic AI-generated images and video you encounter online
Check the privacy settings of any AI tool you use regularly — it takes five minutes

Try This — Exercise 1 (build the verification habit)

Open any AI tool. Ask it for a piece of information you can fact-check — a historical date, a statistic, a claim about something in your industry or field. Then verify it using a search engine or a trusted source. Do this three times in a row.

The goal isn't to catch the AI being wrong (though you might). It's to build the instinct of verifying before trusting as a natural part of how you use these tools — the same way you'd double-check a figure before putting it in a report.

Exercise 2 — Check your privacy settings right now

Open the settings menu in whichever AI tool you use most. Look for:

Whether your conversations are used to train the model (and how to opt out)
Whether chat history is saved, and for how long
Whether there's an option to delete past conversations

Where to find these settings: In ChatGPT → Settings → Data Controls. In Claude → Settings → Privacy. In Gemini → your Google Account → Data & Privacy → Web & App Activity. Most people never do this. It takes five minutes.

Exercise 3 — Practice deepfake scepticism

Next time you see a striking image or short video clip on social media — especially one designed to provoke a strong reaction — pause before sharing it. Ask yourself:

Can I find this same image or video reported by a trusted news outlet?
Does anything look slightly off — hands, teeth, backgrounds, lighting, shadows?
Is there a reason someone would want this particular image to spread?

You don't need special detection tools. You just need the habit of pausing before amplifying. That habit is worth more than any software.

The risks that get talked about vs. the ones that matter now

Public conversation about AI risks tends to cluster around two extremes: the dismissive ("it's just autocomplete, nothing to worry about") and the catastrophic ("it will eventually destroy civilization"). Neither is useful for understanding how to use these tools responsibly today.

The risks that matter most to real users right now are: misinformation spread — including inadvertently by people who trusted AI output without checking it — privacy erosion through careless data sharing, over-reliance on unreliable outputs, and the gradual erosion of critical thinking when we stop questioning what AI tells us. These are real, they're happening now, and they're entirely manageable with simple habits.

What AI companies say about safety — and what that means in practice

The major AI labs — Anthropic, OpenAI, Google DeepMind — all publish publicly about safety. Anthropic in particular has made safety a core part of their positioning; their research on AI behavior and their published model guidelines are unusually transparent. That said, "safety research" at the company level is not the same as "the tool is safe to use carelessly." The practical rules in this lesson apply regardless of what any company says about their product — because the risks covered here are about how people use tools, not how tools are built.

💡

The main risks are practical, not existential. Protect your privacy, verify facts, and maintain your own judgment. AI is a tool — you're still the one responsible for how you use it.

Lesson 07

What to Explore Next

By the end of this lesson, you'll have a few concrete directions to go deeper — without feeling obligated to learn everything.

You already know enough to use AI well. Most people get the majority of the value from what you've covered here. But if you want to go further, pick the direction below that matches what you care about — and follow just that one thread.

If you want to get better at using AI day-to-day

Focus entirely on prompting. Practice writing more context-rich prompts every time you use any tool. Give AI more information than feels strictly necessary and see what changes. The skill compounds quickly with real use.

If you work with data or spreadsheets

AI is surprisingly powerful for structured data. ChatGPT can write Excel or Google Sheets formulas from plain-English descriptions, explain what existing formulas do, analyze uploaded CSV files, and spot patterns in data. Try uploading a real spreadsheet and asking a question about it in plain English.

If you're interested in creative work

Adobe Firefly and Flow by Google are free starting points for images. Midjourney is where you go when you're ready for a step-change in quality. For writing, use AI as a brainstorming partner and structural editor rather than a ghostwriter — it's more useful and more honest.

If you want to follow what's happening in AI

The Verge — AI section Accessible, well-written coverage without requiring technical knowledge. Good for staying current. → theverge.com/ai
MIT Technology Review More in-depth analysis of implications, not just announcements. Worth reading for context. → technologyreview.com
Anthropic's research blog The company behind Claude publishes unusually clear writing on how their models work and how they think about responsible AI. → anthropic.com/research

If you want to go technically deeper

Andrej Karpathy's YouTube channel is the best free resource for understanding how language models actually work — aimed at curious non-experts. Google's "Prompting Essentials" course is free, practical, and takes an afternoon.

Final challenge

Before you close this page, open an AI tool and do one thing you haven't tried before. Not a test prompt — something you actually need. A meal plan, a difficult email, an explanation of something you've been confused about, a plan for a project you've been avoiding.

The best way to learn any tool is to use it for something real.

AI applications that tend to surprise people

Writing difficult personal messages — condolences, apologies, difficult family conversations. AI can help you find the words when you genuinely can't.
Debugging your own thinking — paste in an argument or plan you're unsure about and ask: "What am I missing?" or "What's the strongest counter-argument to this position?"
Learning a new skill fast — ask AI to act as a tutor. Have it explain, quiz you, correct you, and re-explain in a different way until something clicks. Adjust the level as you go.
Understanding a document you've been avoiding — a lease, an insurance policy, a legal letter. Paste it in, ask for a plain-English summary, and ask it to flag anything you should pay close attention to before signing.
Interview preparation — paste in a job description and your CV, then ask the AI to run a mock interview: ask hard questions, push back on vague answers, and give honest feedback. Uncomfortable but useful.
Making sense of a medical situation — not as a replacement for your doctor, but to understand what a diagnosis, test result, or treatment option actually means in plain language before your next appointment. Go in prepared.
Learning another language — ask it to have a simple conversation with you in the language you're learning, correct your mistakes gently, and explain the grammar behind each correction. Cheaper and more patient than most tutors.

💡

You already know enough to start. The rest comes from practice, not more reading. Pick one direction, go a little deeper, and see where it leads.