Series
The Intelligence Industry
A history of neural networks, from 1943 to today, told in episodes.
- Ongoing
- 6 episodes
- 125 min read
This series follows, in the order they happened, the inventions that led from the first artificial neuron to today's large language models. Each episode starts from a concrete problem, shows how it was solved and what the solution cost. Nothing in it goes unsourced.
-
I Didn’t Discover AI Through ChatGPT
A robotics engineer, I spent twelve years alongside artificial intelligence in places where it cannot be separated from a circuit board and a computing budget. This…
-
February 2012: A Car That Learned Nothing
In February 2012, a car drove itself down a Californian highway without having learned anything: every rule had been written by hand. This first chapter shows…
-
From One Neuron to a Network: What Is Learned and What Flows
A neuron on its own decides very little. This chapter stacks neurons into layers, shows that only numbers ever flow through them, counts the weights of…
-
Rows of Numbers: What We Gain and What We Lose
How a network stores and multiplies its numbers, why that makes it a pattern-spotting machine, what it gains from this, what it loses, as the wolves…
-
Rosenblatt, and the Book That Buried His Perceptron
In 1957, in an aeronautical laboratory in New York State, a machine learns instead of being programmed. This chapter tells the story of the perceptron, the…
-
Backpropagation, or the Art of Sharing the Blame
A deep network gets it wrong: how do you know which setting to correct, and by how much? The answer has a name, backpropagation. This chapter…






