Article · Rules and responsibility
An AI playbook somebody actually reads
Nine points, a three-basket model, and a list of what not to put in. The first version takes a day.
Right now
Latest stories- NumbersInfrastructure12 minWhat one AI question actually usesThe bottle-of-water claim comes from a real study whose result changed meaning as it travelled. What it measured, what production measurements now show, and why two honest numbers can differ by a factor of a hundred.Read the article →
- ViewpointApplications20 minVibe coding and AI-assisted programmingTwo different ways of working get grouped under one label, though the distinction that matters is whether the generated code is reviewed. What each approach is good for, what remains after the demo, and what productivity research actually shows.Read the article →
- SurveyApplications18 minAI use cases in plain languageThirty concrete use cases, from the home to the forest, from the production line to a sports club. What the machine does in each, where it goes quietly wrong, and what happens to a use case after it is chosen.Read the article →
- CaseApplications19 minAgents, orchestration and use casesWhat a multi-agent system actually is, the four structures you find in one, what each is good for, and when a single agent is enough.Read the article →
- ViewpointRules and responsibility5 minAI ethics without the sermonLists of principles settle nothing. Seven places where the harm actually happens, and the questions that turn them into decisions.Read the article →
AI: what it is, what it does.
Without jargon.
Explanations, worked examples and current reporting in one calm reference. What the machine actually does, what it costs to run, and where the money ends up.
The ecosystem, layer by layer
Published earlier
All articles →
Article · Infrastructure
Finland's data centre boom and the queue for the grid
Over three gigawatts under contract and 50,000 megawatts as enquiries. What the numbers actually mean, and where the real bottleneck is.
Article · Business
Is there an AI bubble
Four companies will invest 760 billion dollars this year. Meanwhile most companies get no measurable effect on their results. Both can be true.
Start here
Three steps if AI is new to you
The order is deliberate. Fundamentals first, then the whole picture, then the examples.
01
What AI is
Five things that are enough to begin. No technical background, no unexplained terms.
02
What it is made of
The whole chain from electricity and chips to data, models, applications and money.
03
Where it is used
Everyday situations taken apart: what the machine does and what is left to people.
Topics
The parts the whole picture is made of
The fundamentals, five layers from electricity to business, and the rules that cut across all of them.
Concepts
Fundamentals
AI, machine learning and language models. Which one sits inside which.
Layer 1
Infrastructure
Electricity, chips and very large buildings. Everything starts here.
Layer 2
Data
What the machine learns from, and why your own records are the asset.
Layer 3
Models
Training is paid for once. Running it is paid for every single time.
Layer 4
Applications
How a model becomes something that is actually of use to somebody.
Layer 5
Business
Who does what, who pays, and where the money finally settles.
Cross-cutting
Rules and responsibility
What the law requires, what ethics requires, and what has to be written down.
Try it yourself
There is a working language model on this site
Pick a word and the model shows which words are most likely to follow it. Let it keep going on its own and you get sentences that are grammatically fine and mean nothing at all. That is hallucination in miniature.
Open the demo“company builds the car the controls and the material is too slow an expensive chip sits and waits”
Examples
What the machine actually does
“AI handles customer service” tells you nothing. These situations are taken apart piece by piece.
Glossary
66 terms
When a word turns up that you do not know, the explanation should be right here.
- Hallucination
- The model produces a convincing but false statement.
- Inference
- Using a finished model. Every question and answer consumes computation, and therefore money.
- RAG
- The model looks the facts up before answering, so the answer can show its source.
- Token
- Text chopped into small pieces. The model reads and writes tokens, and they are what you are billed for.
Author
This site is written by Harri Salomaa. Forty years in software, twenty of them in the United States and Germany: from collecting process data and analysing network data to immersive computing, and most recently AI. It is written to be read without a technical background, but without simplifying things into being wrong: when something is uncertain, it says so.