counting
Before learning, machines counted.
And it worked.
learning
counting stopped being enough, it had to learn.
attention
learning wasn't enough, it had to choose what to look at.
hello again.
you already knew the ending.
now you know where it came from.
before i go,
there's someone who wants to tell you
why he made all this.
i'll leave you with him.
you already know the ending.
you're missing the path.
you start with a machine that barely knows how to put letters together.you end up building the ideas that make today's machines work.
each one solved something the one before couldn't.
All of this started because of my parents.
I keep meeting more people for whom artificial intelligence is no longer one more tool.
It's the tool they use most during the day.
To work, to write, to learn, to ask, to sort out anything at all.
And yet very few of them could explain what actually happens when they type something and one of these machines answers.
My parents are two of those people.
They don't know about artificial intelligence.
Or about programming.
And there's no reason they should.
So one day I decided to try to explain it to them.
It was going to take one afternoon.
One afternoon wasn't enough.
I started by counting letters. Then came the n-grams, the neural networks, the transformers…
…and that little project ended up turning into this.
None of these machines appeared out of nowhere.
Each one solved something the one before couldn't hold.
That's why, even though you'll probably want to jump straight to GPT or Transformer, I'd start at the beginning.
you can skip to any of them.You can go to whichever chapter you like. But when you understand the problem first, the solution stops looking like magic.
And that's the whole idea.
That someone with no background in computing or maths can go from a machine that barely joins letters all the way to today's LLMs and agent systems, and understand what changes between them.
Not just by reading about it.
By building them.
You'll train your own models, play with them, break them, and get small experiments for understanding what is going on inside.
This is still under construction.
still being built. some chapters are in beta.Some chapters are finished, others are in beta, and a few still don't have the level of detail I want. Transformer, for instance, is quite far along, but it isn't yet the chapter I want it to end up being.
And this page will probably keep growing for a good while.
If it's any use to you, share it.
But above all pass it to someone who doesn't work in this. Someone who uses these machines every day without quite knowing what happens inside.
Because I don't think all of us need to know how to build an AI.
But I do think understanding why it works, why it fails, and what is really behind it helps us use it better.
Understanding them is also learning to use them better.
Thank you for making it this far.
— Adri
well then.
shall we start?