An Overview of Deep Learning for Curious People

There's a reason every deep learning explainer eventually circles back to March 2016. Lee Sedol wasn't some journeyman player picked for a publicity stunt — he held nine dan, the highest rank in Go, and had a trophy case full of world titles to prove it. When AlphaGo beat him 4 games to 1, it wasn't just a scoreline. It was a signal that something had shifted in what machines could actually do.

Go had long been treated as the game computers couldn't crack, and for good reason. Chess has a lot of possible positions, sure, but Go's board and rule set produce a branching complexity that dwarfs it — enough that brute-force search, the trick that let Deep Blue beat Kasparov back in 1997, simply doesn't scale here. For decades, that made Go a kind of unofficial benchmark for genuine machine intelligence, one many assumed was a decade or more away from falling.

Then it fell, in a single week, on a livestream. And the reaction wasn't confined to Go enthusiasts or AI researchers. Lee Sedol's loss became the story that dragged deep learning out of academic papers and into dinner-table conversation. Suddenly people who'd never heard of a neural network were asking what exactly AlphaGo was doing differently.

Weng's piece uses that moment as the entry point into deep learning itself — not because AlphaGo explains everything, but because it's the event that made the question worth asking for a general audience. That's arguably the real legacy of 2016: not the specific techniques inside AlphaGo, but the fact that a five-game match in Seoul became the moment AI stopped being a sci-fi abstraction and started being something people felt they needed to understand.