I Read Every Major CS Paper of the Last 100 Years — So What?

Ever wonder where the tools you use every single day actually came from? Not the marketing story — the real origin. Fireship went and did something a little insane: he traced a hundred years of computer science through ten landmark papers. From Alan Turing in 1936 all the way to GPT-3. And what falls out of that timeline isn't a neat, straight line of progress. It's messier — and honestly, way more interesting.

Let me walk you through the ten, and then I'll tell you the three patterns that jump out when you see them all lined up.

We start in 1936, with Turing's "On Computable Numbers." This is the one that asked the deepest question: what can a machine actually compute, and what can it never compute? He gave us a theoretical model — the Turing machine — and that model is the ancestor of every computer that has ever existed.

Then 1948, Claude Shannon, "A Mathematical Theory of Communication." Shannon invented the bit. He defined entropy. He basically created information theory out of thin air. Every file you've ever compressed, every video you stream — it's all running on Shannon's math.

1958, Frank Rosenblatt, "The Perceptron." The first algorithm that could learn from examples by imitating a neuron. This kicked off the very first wave of AI hype.

And then — 1969 — Minsky and Papert wrote a paper literally called "Perceptrons," and they mathematically proved the thing couldn't handle nonlinear problems. Couldn't even solve XOR. That paper nearly killed neural network research for a whole decade.

1978, Leslie Lamport, "Time, Clocks, and the Ordering of Events." This one's quieter but enormous. Lamport solved the problem of causality in distributed systems — how do you agree on what happened first when there's no shared clock? Logical clocks. Every distributed database you touch today runs on this idea.

1986 — Rumelhart, Hinton, and Williams — backpropagation. This is the resurrection. It brought multi-layer neural networks back from the dead that Minsky had buried. And Hinton? He won the 2024 Nobel Prize in Physics for this line of work.

1998, Brin and Page, "The Anatomy of a Large-Scale Hypertextual Web Search Engine." That's PageRank. They treated the entire web as a graph. That paper is Google.

2012 — AlexNet. Krizhevsky, Sutskever, and Hinton took deep convolutional networks, trained them on GPUs, and cut the ImageNet error rate from twenty-six percent down to fifteen. That was the moment deep learning stopped being a research curiosity and became the industry standard overnight.

2017 — "Attention Is All You Need." The Transformer. They threw out recurrent networks entirely and put self-attention at the center. This is now the backbone of basically every large model you've heard of.

And 2020, GPT-3 — "Language Models are Few-Shot Learners." This showed that if you just scale up and prompt cleverly, you don't need to fine-tune for every task. That's the paper that launched the whole era of prompt engineering.

Okay. Ten papers. Now here's what's fascinating when you zoom out.

First pattern: negation is progress. Minsky demolished the perceptron — and that demolition is exactly what forced the field to think harder, which eventually produced backpropagation. Sometimes being definitively, rigorously wrong is the most useful thing a field can do.

Second: the math shows up before the hardware. Shannon's information theory came decades before the internet. Turing's machine came before real computers existed. The ideas were just sitting there, fully formed, waiting for the infrastructure to catch up.

Third: recombination beats invention. GPT-3 didn't introduce one genuinely new concept. It combined the Transformer, massive datasets, and scaling laws. The breakthrough was recognizing that the combination worked.

So what do you actually do with this? You don't need to go read all ten original papers. But knowing their shape gives you a filter. The next time someone tells you a new framework "changes everything," you've got a real question to ask: is this a genuine paradigm shift — or is it an old idea that's just now waiting for the hardware to catch up?

Three things to hold onto. One: progress in CS is not linear — ideas get buried, rediscovered, and revived, sometimes decades apart. Two: the theory almost always arrives before the machines that make it useful. And three: the biggest leaps often aren't new inventions at all — they're someone noticing that pieces we already had fit together. Keep that lens, and you'll read the next hype cycle a lot more clearly.