Skip to content
LearnWithZavi
All lessons

The Big Picture ยท Lesson 2

A Very Short History of AI

From perceptrons to ChatGPT in one caffeinated lesson.

Step 1 of 7: Seventy Years in Five Minutes

Seventy Years in Five Minutes

AI's history reads like a startup story: bold promises, spectacular faceplants, long stretches of nobody caring, and then, suddenly, everybody caring at once. Grab your coffee. We're doing seven decades at highway speed.

EraWhat happenedMood
1950s to 60sThe perceptron, a simple artificial neuron that could learn basic patterns. Newspapers promised thinking machines "soon."Wild optimism
1970sFirst AI winter. Perceptrons hit hard limits, funding froze.Existential shivering
1980sExpert systems (giant piles of hand-written IF-THEN rules) boom...Cautious hype
Late 80s to 90s...and bust. Second AI winter.Sweater weather again
2012Deep learning breaks through: neural networks with many layers, fed by big data and gaming GPUs, crush an image-recognition contest."Wait, it works now?!"
2017The transformer arrives via a paper titled "Attention Is All You Need."Quiet revolution
Nov 2022ChatGPT launches. 100 million users faster than basically any product ever.Absolute pandemonium

The Perceptron: A Neuron With Big Dreams

In the late 1950s, researchers built the perceptron, a mathematical imitation of a single brain cell. It could learn to tell simple shapes apart, and the press went feral, predicting machines that would walk, talk, and be conscious any day now.

Then someone proved perceptrons couldn't handle certain embarrassingly simple problems. Funding evaporated. Welcome to the first AI winter, a period when "I work in AI" was a great way to end conversations at parties.

AI winters are named like actual winters because research funding, like a bear, went into hibernation. Twice.

Quick check

The perceptron story follows a pattern. Which one?

The Thaw: Deep Learning

The core idea never died: stack many layers of artificial neurons and let them learn from examples. What was missing was fuel: enough data (hello, internet) and enough compute (hello, graphics cards built for video games).

By 2012, both had arrived. A deep neural network demolished the competition in a famous image-recognition challenge, and the field flipped from "quaint academic hobby" to "arms race" almost overnight.

Quick check

Stacking layers of artificial neurons was an old idea. What changed by 2012 to make it work?

2017: Attention Is All You Need

In 2017, researchers published a paper with the most confident title in computer science: "Attention Is All You Need." It introduced the transformer, an architecture that processes language by letting every word pay attention to every other word in a sentence, all at once, instead of trudging through one word at a time.

Transformers turned out to scale beautifully. Bigger models plus more data kept producing better results, like a cheat code that never got patched. Nearly every famous model today (GPT, Claude, Gemini) is a descendant of that one paper. The "T" in GPT literally stands for Transformer.

Quick check

What did the transformer do differently with language?

November 2022: The ChatGPT Moment

Large language models had existed for years, but mostly behind APIs and research demos. Then in November 2022, OpenAI wrapped one in a friendly chat box and let the public in.

The reaction was seismic. Suddenly your uncle, your dentist, and your group chat all had opinions about AI. Decades of research compressed into one cultural moment: the "iPhone moment" of artificial intelligence, minus the queue outside the store.

The lesson from history: AI progress isn't a smooth ramp. It's long plateaus punctuated by sudden cliffs (in the good direction). Which is exactly why the field is so hard to predict.

Quick check

Large language models existed before November 2022. What was new about ChatGPT?

Checkpoint

Checkpoint

AI went perceptrons, two winters, deep learning, transformers in 2017, and the ChatGPT boom in November 2022.

Timeline Quiz

๐Ÿ“ Quiz

1 of 4

Swipe right if this answers the question correctly, left if it does not.

What is an "AI winter"?

โ€œThe codename for the first neural networkโ€

Is this the right answer?

Found this useful? Pass it on.