Korva is a scroll of the internet’s best ideas.
Humans can't feel exponentials
Linear instincts
Picture foraging for food: a good hunt brings a couple of days of meals. On an exponential curve, the equivalent of a good hunt could be a thousand years of food, and nothing in the brain gives an intuitive signal for that. The same curve shows up in AI's progress, in the scaling laws, and in the revenue of the companies building it.
A lesson from his parents
The lesson came early to Scott Wu from his parents, who grew up in communist China in the 1960s, much poorer and in a very different life. They arrived in the US to find everyone with a car and all kinds of household appliances, a change they found jarring, and it was slow compared with the pace of AI today.
Forgetting life before
People got used to all of it fast, and now can't live without it. The same pattern is likely with AI: five years from now it will be insane to think about what people will have, and ten years from now they will have forgotten ever living without it.
Why do brains misjudge exponential growth?
Human intuition developed in settings where quantities changed by steady amounts, such as distance walked or food gathered. In 1975, psychologists Willem Wagenaar and Gideon Sagaria asked people to extrapolate exponential series and found that they consistently guessed far too low. The error appears even in people who know the math. It tends to grow larger as the number of steps increases.
How do AI scaling laws predict progress?
In 2020, OpenAI researchers led by Jared Kaplan found that a language model's error drops in a smooth, predictable curve as model size, data, and compute grow. The relationship held across several orders of magnitude. Labs use those curves to estimate how a larger model will perform before they train it. The pattern is a power law, which is why progress can look sudden when each doubling of compute is plotted on a normal scale.
How quickly did American households get cars?
Henry Ford introduced the Model T in 1908. By 1930, more than half of American households owned an automobile, according to historical estimates built from census and registration data. The adoption curve was steep enough that roads, gas stations, and suburbs reshaped around it within a generation.
More from David Senra
