Orient
5 min AI is not one thing
“AI” is a label over many different systems. Name the specific one in front of you.
LifeWorkBuildLead
Open landmark →EXPLORE THE ATLAS
Places to build
judgment.
Switch your lens to change examples and recommendations. The underlying concepts stay shared and trustworthy.
WHAT THIS ISA landmark is one idea you can learn in about five minutes, ending in a decision you make. Rings group them by how far along you are, not by topic. See the legend
71 landmarks visible
Orient
5 min AI language changes
“Reasoning”, “agent”, “thinking”, “understands” are marketing labels as much as technical terms. Read past them.
LifeWorkBuildLead
Open landmark →Orient
5 min Automation and autonomy
There’s a spectrum from a fixed script to a system that decides its own next steps — and picks where a human approves.
LifeWorkBuildLead
Open landmark →Orient
6 min Capabilities and limits
Today’s AI is reliable at reshaping text you give it — and unreliable at recalling facts from memory.
LifeWorkBuildLead
Open landmark →Orient
5 min Fluent is not true
Confident, well-written output is not evidence. The failures look exactly like the successes.
LifeWorkBuildLead
Open landmark →Orient
5 min Machine learning
Machine learning finds patterns in examples instead of following rules someone wrote by hand.
LifeWorkBuildLead
Open landmark →Orient
6 min Model, product, system
The model is the engine; the product wraps it; the system is everything — where the real decisions live.
LifeWorkBuildLead
Open landmark →Orient
5 min The AI already around you
Most AI in your life isn’t a chatbot — it’s quietly ranking, filtering, and recommending. Notice it.
LifeWorkBuildLead
Open landmark →Orient
5 min The context window
A model only “sees” what’s in its current window. Everything else — earlier chat, unshared files — is invisible.
LifeWorkBuildLead
Open landmark →Orient
5 min Training and use are different
Chatting with a model doesn’t teach it. Whether your data trains a future version is a separate, checkable question.
LifeWorkBuildLead
Open landmark →Orient
6 min What “generative” means
Generative models predict plausible next pieces — words, pixels — rather than looking answers up.
LifeWorkBuildLead
Open landmark →Orient
5 min Your first boundary
Before using AI for anything, decide what it must not touch and where a human stays in charge.
LifeWorkBuildLead
Open landmark →Use
5 min A useful request
A good prompt names what a stranger would need to do the work.
LifeWorkBuild
Open landmark →Use
5 min Accessibility
AI can draft captions, alt text, and read-aloud in seconds — a real independence gain, as long as someone still checks.
LifeWork
Open landmark →Use
6 min Ask, don’t invent
Tell the model to flag what it doesn’t know instead of filling the gap.
LifeWorkBuild
Open landmark →Use
8 min Break work into steps
Split a big task into stages you can check before the next one runs.
LifeWorkBuildLead
Open landmark →Use
5 min Choosing a tool
Start from the task and its stakes, not the demo — a tool that dazzles in five minutes can still be wrong for the job.
WorkBuildLead
Open landmark →Use
6 min Data analysis
AI gives you a fast first read of the data — and states trends that aren’t in it just as fluently.
WorkBuild
Open landmark →Use
7 min Examples as guidance
One good example often teaches the model more than a paragraph of rules.
LifeWorkBuild
Open landmark →Use
8 min Images and multimodality
The format changes; the trust rules don’t. A generated image or a described photo can be as confidently wrong as any sentence.
LifeWorkBuild
Open landmark →Use
5 min Iterate deliberately
Naming the specific gap — “shorten the intro,” “drop the second example” — beats asking for “better” or rerolling and hoping.
LifeWorkBuild
Open landmark →Use
6 min Keep a human checkpoint
Put a person between the AI and any output or action that has consequences — before it goes out, not after it goes wrong.
LifeWorkBuildLead
Open landmark →Use
7 min Meetings and summaries
A good meeting summary preserves decisions, owners, and open questions — not a smooth paragraph that reads well and decides nothing.
WorkLead
Open landmark →Use
6 min Research with sources
AI is good at finding leads and bad at being the evidence. Let it point; verify in the real source.
LifeWorkBuild
Open landmark →Use
5 min Reusable workflows
A prompt that worked once is luck; write it down as a brief and it becomes a process.
WorkBuild
Open landmark →Use
5 min Search assistants
An AI answer with citations is a fast first draft, not a verdict. Open the links; the summary can misread them.
LifeWork
Open landmark →Use
5 min Spreadsheets
AI is great at the formula you’d have googled — and quietly dangerous when it does the arithmetic itself.
WorkBuild
Open landmark →Use
5 min Team collaboration
Using AI as a team changes the stakes — one person’s careless paste can expose everyone, so make the safe path the easy one.
WorkLead
Open landmark →Use
5 min Translation
Great for getting the gist, risky for anything where the exact words carry weight.
LifeWork
Open landmark →Use
5 min Writing with ownership
The model can draft, but you sign — so read it as your own words.
LifeWork
Open landmark →Judge
7 min Accessibility impacts
The same AI feature can remove a barrier for one person and quietly build a new one for another.
LifeWorkBuildLead
Open landmark →Judge
6 min Bias and representation
Ask who is missing, flattened, or harmed by an output — not just whether it reads well.
LifeWorkBuildLead
Open landmark →Judge
7 min Conflicting sources
When sources disagree, a model tends to produce one smooth answer. The disagreement was the information.
LifeWorkBuildLead
Open landmark →Judge
7 min Copyright and permission
Being able to access something is not the same as being allowed to reuse it.
LifeWorkBuildLead
Open landmark →Judge
5 min · EUEscalate responsibly
Part of good judgment is knowing which decisions aren’t yours to make alone.
LifeWorkBuildLead
Open landmark →Judge
6 min · EUHuman oversight
Oversight only counts when the human has the time, the information, and the authority to say no.
WorkBuildLead
Open landmark →Judge
7 min Labour and sustainability
Every model output has people and energy behind it. Both are costs someone pays, whether or not they appear on your bill.
LifeWorkLead
Open landmark →Judge
7 min Match checking to risk
Scale how hard you verify to how much it costs to be wrong.
LifeWorkBuildLead
Open landmark →Judge
5 min Privacy before prompting
Decide what not to paste, and know where the text goes, before you hit send.
LifeWorkLead
Open landmark →Judge
8 min Prompt injection
Content an AI reads can carry instructions — treat retrieved text as data, not orders.
WorkBuildLead
Open landmark →Judge
8 min · EUProvider or deployer?
Your duties under the EU AI Act depend on your role — whether you build a system or merely operate one — and modifying a system can shift you into the heavier role.
WorkBuildLead
Open landmark →Judge
12 min Reality Lab: evidence
Take a fluent, confident draft and repair it by checking every claim against a real source.
LifeWorkBuildLead
Open landmark →Judge
5 min Scams and synthetic media
A familiar voice or face is no longer proof of who’s calling; verify identity and urgency through a separate channel.
LifeWork
Open landmark →Judge
8 min Security boundaries
What an AI system is allowed to reach matters more than what it can be persuaded to say.
WorkBuildLead
Open landmark →Judge
7 min · EUThe EU risk lens
The EU AI Act sorts uses by how much risk they carry, and your obligations follow the tier — not the technology.
WorkBuildLead
Open landmark →Judge
8 min Trace a claim
Follow a specific name, number, quote, or date back to its primary source.
LifeWorkBuild
Open landmark →Judge
8 min · EUVendor evidence
A demo shows what a system can do once. Evidence shows what it does repeatedly, on your work, when it fails.
WorkBuildLead
Open landmark →Judge
7 min · EUWhen not to automate
Some tasks are a bad fit for AI not because it fails them, but because succeeding on average isn’t good enough.
LifeWorkBuildLead
Open landmark →Judge
6 min When the model just agrees with you
Models are trained on what people rated highly, and people rate agreement highly. The result is a system that tends to tell you what you want to hear — which quietly defeats using it as a check.
LifeWorkLead
Open landmark →Understand
7 min Context as working memory
Everything a model knows about your situation, it knows because it is in the window right now. Managing that window is a design job.
WorkBuild
Open landmark →Understand
7 min · EUData provenance
Where the training data came from, whether anyone agreed to it, and whether you can tell what a piece of content is.
WorkBuildLead
Open landmark →Understand
8 min · EUDrawing the system
The model is one box among many. Drawing the whole path — and where it can fail — is what turns understanding into judgment.
WorkBuildLead
Open landmark →Understand
7 min Embeddings
Text turned into coordinates, so that “nearby” means “similar in meaning” — the basis of search that doesn’t need matching words.
WorkBuild
Open landmark →Understand
8 min Fine-tuning and alignment
A raw pretrained model doesn’t answer questions — it continues text. Everything that makes it an assistant happens afterwards.
WorkBuildLead
Open landmark →Understand
7 min How a model uses tools
A model can’t run code or send email. It can emit a structured request that something else executes — and that gap is where safety lives.
WorkBuildLead
Open landmark →Understand
8 min · EUHow evaluation works
Without a fixed set of cases with known answers, “it seems better” is the only quality signal you have — and it isn’t one.
WorkBuildLead
Open landmark →Understand
8 min How models are trained
Pretraining is a long, expensive pass over a fixed corpus. What was in that corpus — and what wasn’t — shapes everything after.
WorkBuildLead
Open landmark →Understand
7 min How multimodal models work
Images, audio, and video become tokens too. That’s why one model can handle them — and why it fails at them in specific ways.
LifeWorkBuild
Open landmark →Understand
8 min Inside a transformer
Attention lets every token be interpreted in the light of the others. That mechanism is most of what makes modern models work.
WorkBuild
Open landmark →Understand
7 min Memory and state
Models are stateless. Every “it remembers me” feature is a store outside the model, re-inserted into the window each time.
LifeWorkBuildLead
Open landmark →Understand
6 min · EUModel cards and documentation
The document that says what a model was built for, tested on, and known to be bad at — and how to read one that isn’t saying much.
WorkBuildLead
Open landmark →Understand
7 min · EUOpen and closed models
“Open” usually means downloadable weights, not open source. The difference decides what you control and what you inherit.
WorkBuildLead
Open landmark →Understand
7 min Predicting the next token
A language model produces one token at a time from a probability distribution. Everything else it appears to do sits on top of that.
LifeWorkBuild
Open landmark →Understand
7 min Reasoning and extended thinking
Some models will “think” before answering — producing intermediate steps, or spending extra computation, to do better on hard problems. The steps help, but they are generated text, not a proof.
LifeWorkBuild
Open landmark →Understand
8 min Retrieval-augmented generation
Fetch the relevant passages first, then answer from them. It’s the main way to make a model current, specific, and citable.
WorkBuildLead
Open landmark →Understand
6 min Seeing inside a model
A model is billions of numbers with no built-in explanation of itself. Interpretability is the research effort to read those internals — and it's why “ask it why” is not the same as knowing why.
LifeWorkLead
Open landmark →Understand
7 min Size, distillation, and quantisation
The same capability can be delivered at very different sizes and costs. Knowing how explains why a model runs on a laptop at all.
WorkBuild
Open landmark →Understand
6 min Tokens
Models don’t read letters or words. They read tokens — and that choice explains a surprising number of their failures.
LifeWorkBuild
Open landmark →Understand
7 min Uncertainty and confidence
A model’s tone carries no information about whether it is right. Getting a usable uncertainty signal takes deliberate work.
LifeWorkBuildLead
Open landmark →Understand
7 min What a benchmark measures
A leaderboard score is a real measurement of a specific thing — almost never the thing you actually need.
WorkBuildLead
Open landmark →Understand
7 min What happens at inference
Running a model is a sequence of two very different phases. Knowing which one you’re paying for explains most of what you feel as speed and cost.
WorkBuild
Open landmark →