The Teacher, Part 3: 2012, and the students who changed the world

A scientist’s citations are not the measure of a teacher. The measure of a teacher is the people who walk out of the room better than they walked in, and what they go on to do. By that measure Geoffrey Hinton’s single most important year may not have been any of his own discoveries. It was 2012, and the work was done by two of his students.

The margin nobody had seen

For years the standard test of computer vision was the ImageNet competition: show a program a million-plus labelled photographs, then see how well it can name what is in images it has never seen. Progress was incremental, a percentage point at a time, and neural networks were a fringe idea almost nobody in the field took seriously.

In 2012, two graduate students in Hinton’s lab at the University of Toronto — Alex Krizhevsky and Ilya Sutskever, with Hinton as their advisor — entered a deep neural network of the unfashionable kind. It was trained on a pair of consumer gaming graphics cards in a bedroom, a detail that still delights people who know how little that is. It did not edge ahead of the field. It beat every other research lab by a margin of more than ten percentage points — a crushing result in a contest usually decided by fractions. The network was named AlexNet, after Krizhevsky who wrote it.

That was the moment the dam broke. Before AlexNet, very few machine-learning researchers used neural networks; after it, almost everyone did. The paper — “ImageNet Classification with Deep Convolutional Neural Networks,” by Krizhevsky, Sutskever and Hinton — has since been cited well over a hundred and fifty thousand times, among the most cited scientific papers ever written. In 2024 the Computer History Museum preserved the original AlexNet source code as a historical artifact. The modern era of artificial intelligence has a birthday, and it is 2012.

What Hinton says about his own role

Here is the sentence we quoted in Part 1, and it deserves a part of its own. Describing the project for the Computer History Museum, Hinton put the division of labour this way: “Ilya thought we should do it, Alex made it work, and I got the Nobel Prize.”

It is a joke, and it is also the truest thing a teacher can say. The idea came from one student. The execution came from another. The advisor created the room in which two young people could chase an unfashionable hunch with real resources and academic freedom — and then he said, in public, at the peak of his fame, that the credit runs the other way. There is a well-told account that when the three of them later divided a large sum from the company their work spun out, the two students insisted their professor take the larger share, over his objections. That is what the relationship was.

And look at where the students went. One of them, Ilya Sutskever, went on to co-found and lead the science at one of the most influential artificial-intelligence labs in the world — the lab behind the systems that put “AI” in front of hundreds of millions of ordinary people. Trace the line: an unfashionable idea, a professor who kept the faith when the field had moved on, two students he trained, one gaming-card experiment in 2012 — and from it, most of what the public now means by the word intelligence in a machine.

The lesson we actually take from this

We run a learning mission, so of all the parts of Hinton’s story this is the one we cannot look away from. His largest contribution to humanity may be transitive. He taught people, and the people changed the world. The science mattered enormously, but the multiplication happened through students.

That is, almost exactly, the premise of Meta DAX. The oldest constraint in education is the ratio — too few teachers, too many learners, and the best teaching rationed to whoever can pay for proximity to it. Our whole bet is that this ratio can be inverted: that teaching can be made abundant rather than scarce, handed to every human being free and forever, so that the thing that multiplied through one Toronto lab can multiply through anyone, anywhere, who has something to learn or something to teach. The cognition behind our tools is provided by Obsidian Delta; the shape of what we are trying to do with it — get the best out of every single person, and let their best pass on to the next — is written right here, in a story about two students and a professor who said they deserved the credit.

A teacher is a multiplier. 2012 is the proof. Everything we are building is an attempt to make that multiplier available to people who will never have a Toronto lab of their own.

This is Part 3 of five. Next: the turn — when the teacher, having built all of this, began to warn the world about it.

---

This is a tribute and our own commentary. The DAX Foundation and Obsidian Delta are not affiliated with, endorsed by, or in partnership with Geoffrey Hinton, the University of Toronto, the Computer History Museum, or any person named here. Every quote is attributed to its public source; nothing is put in anyone’s mouth.