Here’s how the greats make decisions: they commit fully when the information is high signal, and experiment widely when it’s noisy. In 2014, Diederik Kingma and Jimmy Ba taught a machine to do the same thing. Their paper, “Adam: A Method for Stochastic Optimization,” described an algorithm that adjusts each weight in a neural network with small nudges when the signal is noisy and big ones when it’s clean. Adam outperformed every other optimizer of its time and became the default for training neural networks.

I’ve studied the greats. I’ve read hundreds of biographies, strategy books, and shareholder letters for fun (or because I’m insanely obsessed, but what’s the difference?). Here’s how they used Nudges and Leaps in their decision making to lead the best teams in history.

People Decisions

The best teams hired slow and fired fast. People are noisy by default because they have decades of life and experiences you can never assess in a week or two. Interviews test how someone performs in an interview. Trial projects test how someone performs on a trial project. Neither tells you what happens when unexpected pressure hits, when family life starts trumping work life in a way that hurts morale, when the real shape of the person shows up.

Same thing in personal life. Getting love-bombed by a stranger and falling for them, calling someone your brother after one bro down in Montana, and wanting to marry someone who reminds you of your imaginary ideal after the third date. These are all leaps that should have been nudges.

Signal starts when you find faults you can live with. Nobody is perfect, and perfect is too good to be true. Until you’ve seen someone’s cracks and decided you still want them around, you haven’t nudged enough.

So nudge. Hire slow. Give people real work. Watch them over time. Commit hard only when they prove they can crush the work and make the team stronger.

Idea Decisions

New ideas are a siren song. They seduce you into killing your bank account or years of your youth.

An idea earns commitment through residency. You read widely around it. You talk to people who’ve tried it. You practice it. You adopt it for a while and watch what it does to your decisions, your habits, your work. You hold them loosely and try to disprove them.

The trap is building on ideas you haven’t lived with. Whole strategies get built on a phrase someone read in a book. Whole companies get built on a theory that sounded right in a podcast. Six months later the strategy collapses, because the idea underneath it was noisy and the leap was premature.

Read widely. Practice the idea. Let it live in your life before you build on top of it. Go all out science on that mfer and try to disprove it. Commit to it and build on it only when it has survived real scrutiny. That’s how you collect reinforcing ideas that compound your leverage in every decision you make.

Craft Decisions

The sequential trap kills most makers. “I’ll learn it all, then I’ll make it.” Nobody who has made anything has ever learned it all first.

Preparation and experimentation run in parallel. Set a target then use your own judgment and external feedback to navigate the way. Let’s say you want to draw a photorealistic pencil sketch of my cat. This goal should be met by both your taste standards and the external feedback that makes it true. Now the whole journey calibrates. So you learn how to draw eyes to your satisfaction. Then the nose. Then the fur. Then the depth. Then the lighting. Then you put it all together and show strangers. Feedback has to be part of the process or you can never calibrate your level of craft.

Pick a target. Run tight loops on each component. Integrate. Get feedback from people whose judgment you trust and from complete strangers. Try selling it for the most valuable feedback of them all. And if the result is something you’re willing to run more loops in, commit harder and get better.

Desire

Any optimization process, including Nudge and Leap, can’t work without a desire.

Adam needs a loss function. The algorithm optimizes toward a defined target, and without one, it has nothing to do. The human version is the same. You can’t tell signal from noise without knowing what you’re trying to get to. You can’t size a nudge or a leap without knowing what they’re serving.

Most desires are borrowed. You see someone succeed and want what they have. You read a book and adopt its vision. You scroll a feed and absorb its wants. Running Nudges and Leaps on a borrowed desire gets you to somewhere you don’t actually want to be, efficiently.

The work before the work is making sure the desire is yours. Sit in silence until the noise of other people’s wants fades. Notice what you keep coming back to when nothing is pulling on you. That’s the desire worth optimizing toward.

Without a goal there is no optimization. Without traction there is no distraction.

Thank you for reading this reaction to:

Adam: A Method for Stochastic Optimization — Kingma, Ba — 2014 — Adaptive gradients train everything reliably

Download PDF