There’s a moment before every wave where you have to make a decision. You’ve been watching the set for thirty seconds — the way the horizon lifts, the particular shade of dark blue that means deep water moving fast. Then you paddle. Or you don’t.
I’ve been thinking about this a lot lately, because I’ve spent the last year working with large language models and neural networks, and the analogy keeps surfacing (sorry).
Pattern recognition at scale
Machine learning models don’t understand the world the way we do. They don’t know that a wave is a wave. They find patterns in enormous datasets and learn to associate inputs with outputs. The model that reads a sentence doesn’t comprehend meaning — it predicts what token is likely to follow.
And yet. The outputs can be startlingly apt. Sometimes uncannily so.
Learning to surf works similarly. In your first year, you’re conscious of everything — the angle of your board, where your feet are, where your hands are. By year five, you’ve stopped thinking. Your body has built a model of the ocean: what a close-out looks like, how a beach break differs from a point break, the specific rhythm of a three-wave set.
You don’t decide to read the ocean. You just do.
Feedback loops
The thing that makes neural networks learn is gradient descent: the model makes a prediction, measures how wrong it was, and updates its weights accordingly. Do this millions of times and you get something that seems to understand.
Surfing has the same architecture. Every wipeout is a data point. Every perfect ride updates the model. The ocean doesn’t care about your feelings — it’s pure, dispassionate feedback.
The difference is time. A model can train on a billion examples in a few hours. The ocean gives you maybe ten decent waves per session. Which is why learning to surf takes years, and why it feels like such a precious thing when it finally clicks.
A bit of humility
Here’s what both pursuits have taught me: confidence and humility aren’t opposites. The better you get, the more acutely aware you become of how much you don’t know. The best surfers I’ve met are the most respectful of the sea. The best ML researchers I know are the most careful about what their models can and can’t do.
The ocean is not a problem to be solved. Neither is intelligence.