A pocket reference for AI, in plain English.
Every topic opens with a 30-second gist. If you want more, expand a section. Nothing is forced on you. 78 topics, all written — more added regularly.
Search 78 topics
Or follow a path
A short, ordered set of topics for a specific moment. Built by hand, never auto-generated.
I'm worried about my mum or dad.
Family topics · voice clones · panic call · safe-word · romance scams · setup checklist
I'm just starting to learn about AI.
How AI works · LLM · prompt · training data · hallucination · agents
My kid uses AI for homework.
Kids & school · how AI writes · when to step in · honest use · tutoring
I work in a non-tech job.
Work topics · pasting work · meetings · disclosure · auto-replies · enterprise vs public
How AI works
14 topicsContext window
How much an AI can read in one go — and why it forgets the earliest part of a long conversation.
Deepfake
Deepfake — a photo, video, or voice clip that looks real but was made by a computer. 30-second gist with optional depth.
Embedding
An embedding is how a computer turns words into numbers it can compare. The quiet machinery behind search, recommendations, and 'more like this'.
Fine-tuning
Fine-tuning is teaching a general AI to do one specific job well. Plain-English explanation of when and why companies do it.
Generative AI vs the older kind
Not all AI is ChatGPT. The recommendation engine, the fraud detector, the spam filter — those are AI too, just a different kind.
Hallucination
Hallucination — when an AI says something untrue with full confidence, as if it were a fact. 30-second gist with optional depth.
Large language model
A large language model (LLM) is the engine behind ChatGPT, Claude, and Gemini. Plain-English explanation of what it is and what it can and can't do.
Multimodal AI
Multimodal AI can read text, look at pictures, and listen — at the same time. Plain-English explanation of why this changes everything.
Open weights
An open-weights AI is one anyone can download and run themselves. Plain-English explanation of why this matters.
Prompt
A prompt is what you type to an AI. Why a small change in wording can change the answer a lot — plain-English explanation.
RAG · answering from your documents
How an AI answers from a specific set of documents — your company's policies, your library of PDFs — without retraining the whole model.
Reasoning models
A new generation of AI that pauses to think before answering — slower, more expensive, often more accurate.
Token
A token is the chunk of language an AI works with. Why a long letter sometimes 'costs' more than a short one.
Training data
Training data is the books, websites, and conversations an AI learnt from. Plain-English explanation of what's in it and why it matters.
In the wild
13 topics"AI overview" in search
The little AI summary at the top of Google search results. When to read it, when to skip past, when to mistrust it.
"This email feels off"
Why so many emails suddenly read the same. What it tells you. Plain-English signals of AI-written messages.
AI books on Amazon
Why Amazon's Kindle store is full of AI-generated books in 2026. What to look for. The real harm in some categories.
AI in customer support
When you're talking to a bot, when to ask for a human, and how to tell. Plain-English guide.
AI in social media comments
Bot accounts using AI to comment on Reddit, X, YouTube. What they're trying to do. How to spot them.
AI in your smart speaker
Alexa, Google Home, and Siri moved to LLMs in 2024-25. What changed under the hood. What you can switch off.
AI music in your playlists
How much of Spotify is now AI-generated? And does it matter? Plain-English explainer.
AI on dating apps
Profiles, photos, and the new shape of romance scams. Plain-English explainer.
AI summaries on news sites
The little box at the top of every article. When to trust, when to scroll past. Plain-English guide.
AI-written reviews
Why so many product and restaurant reviews suddenly read the same. How to spot one in 2026.
Face filters and you
What happens to your face after you scan it for a beauty filter. Plain-English privacy guide.
Spotting an AI image
The 5 telltales of an AI-generated image in 2026 — and the 3 that no longer work. Plain-English checklist.
Voice clone
A voice clone is a synthetic copy of someone's voice, made from a short audio sample. How much voice does an AI need? Less than you think.
Trust & truth
11 topics"AI safety" — what people mean
Three different things people mean by 'AI safety'. Plain-English breakdown so the next news story makes sense.
"I'm not sure" — when AI says it
A rare moment that means more than the answer itself. Why AI uncertainty is worth listening to.
AI citations
How to tell a real AI citation from an invented one in under 10 seconds. Plain-English habit.
Cross-checking AI
A 30-second routine for any AI answer that actually matters. Plain-English, practical.
Date awareness
Why AI sometimes thinks it's still 2023, and what to do about it. Plain-English explainer.
Getting AI to admit its limits
A handful of prompts that consistently get more honest, more uncertain answers from AI.
Lying vs confabulation
Why 'the AI lied to me' isn't quite right — and what's actually happening when an AI says something untrue.
Spotting AI-written text
The patterns that give AI-written text away — and why they keep changing.
Sycophancy
Why the AI keeps agreeing with you — and why that's a warning sign, not a compliment.
Watermarking AI content
The hidden signal AI companies are trying to embed in generated images and text — and whether it actually works.
When AI refuses to answer
Why AI sometimes says "I can't help with that". What's a sensible safety policy versus an over-cautious glitch.
Privacy
10 topicsAI in health apps
What happens when health apps get smarter — and what you should know about the data behind them.
AI in your email
When you grant Copilot or Gemini access to your inbox — what does it actually see? Plain-English explainer.
AI in your smart TV
What your smart TV processes — voice search, on-screen ads, automatic content recognition. And what to switch off.
AI screening your CV
Most large employers now use AI to screen CVs before any human reads them. What this means for how you write yours.
Browser extensions
The 'AI assistant' extension you installed last week. What it's reading. Plain-English security check.
Deleting your AI history
What 'delete' actually does on each major AI service. Plain-English clarity.
Photo upload risk
If you uploaded a photo of yourself to an AI app — what happens next? Plain-English privacy primer.
Training opt-outs
The single setting in each AI that says 'don't learn from me'. Plain-English checklist.
What ChatGPT remembers
What's stored, what's used to train, and what you can switch off. Plain-English privacy guide.
When AI leaks
Three real cases where AI gave one user another user's data. Plain-English summary.
Family
10 topicsAI and grief
Apps that recreate a deceased relative from their messages and recordings. What's possible. What's wise.
AI as a tutor for kids
When AI tutoring helps. When it stalls real learning. Plain-English guide for parents.
AI companion apps
Replika, Character.AI, and others — apps you can have a real conversation with. The good, the awkward, the worrying.
AI for wedding speeches and eulogies
When AI helps you write something heartfelt — and when it gets in the way.
AI in kid's homework
When to encourage it, when to step in, when to walk away. Plain-English guidance for parents.
Helping a parent use AI safely
Three settings to change before you walk away. Plain-English checklist for setting up AI for an older relative.
The "panicked relative" call
The voice-clone playbook scammers are using right now — and how to defuse it.
The AI-aided romance scam
Why the messages feel uncannily right. What gives them away. Plain-English warning for older relatives.
The family safe-word
A two-minute conversation that one day saves a lot of money — and a lot of fear.
When a kid sees an AI image
A short conversation that helps a lot. Plain-English script for parents.
Work
11 topics"AI was used" — when to disclose
A short rule that keeps you safe from quiet career damage. Plain-English ethics for AI at work.
"Enterprise" AI vs the public one
Why your IT team really cares which one you use. Plain-English explanation of the difference.
"Will AI replace me?"
The honest answer: not your job — your tasks. Plain-English take on AI and work in 2026.
AI and knowledge work
Lawyers, accountants, teachers, analysts — what AI is changing in 2026 for each. The skill that has gained the most value.
AI auto-replies
When the email you got was written by AI. When yours was. When to stop.
AI in customer-facing roles
When the AI helps. When it sounds like it helps. The difference. Plain-English take.
AI in hiring
Beyond CV screening: AI in interviews, video screening, scheduling, reference-checking. What's actually happening in 2026.
AI in performance reviews
Some companies now use AI to summarise your last year of work — meetings, emails, code commits — for performance review prep.
AI in your meetings
What's recording, who hears it, when to ask the AI to step out. Plain-English meeting etiquette.
AI mistakes that cost real careers
What 2023-25 taught us about how AI quietly damages careers. The pattern is the same every time.
Pasting work into AI
A short rule for what you can paste — and what you really shouldn't. Plain-English work safety.
Looking ahead
9 topics"AGI"
The phrase you keep hearing on the news. What it actually means, in plain English.
AI & energy
How much electricity does a chat actually cost? What you can do. Plain-English.
AI agents
An AI that doesn't just answer — it goes off and does things. What that changes.
AI in elections
What 2024-25 elections taught us about AI and democracy. Less catastrophe than feared, more drift than expected.
AI personhood
Should AI have legal status? The question is starting to creep into real courts. What's the answer in 2026?
AI regulation
The EU AI Act and the rest. What it means for you. Plain-English summary.
New jobs AI is creating
Three real new jobs, three hyped ones. Which is which. Plain-English.
Open vs closed AI
Why the open vs closed AI fight matters and which side does what.
What "AI safety testing" actually involves
Before a new AI ships, what's actually tested? What gets through anyway?