Sublime

that reward maximization really isn’t the whole story of why we do what we do.

want would be some way to gauge whether the situation we are in—as represented by the pixels on the screen—was meaningfully similar to one we’d been in before.

Fads and fashions are the result of following others’ behavior without being anchored to any underlying objective truth about the world.

When balancing favorite experiences and new ones, nothing matters as much as the interval over which we plan to enjoy them.

And it gives a remarkably straightforward solution to the problem of how to combine preexisting beliefs with observed evidence: multiply their probabilities together.

Do the difficult things while they are easy and do the great things while they are small.

you learned two important things: how to take the right action in a given situation, and how to estimate the likely future rewards that a state of affairs might hold.

We want to act in ways that will be understandable to machines, and we also want our machines to act in ways that are “legible” to us.

Given the complexity of the world, why on earth should such dead-simple models—a simple tally of equally weighted attributes—not only work but work better than both human experts and optimal regressions alike?