The shepherd who never searches for lost sheep.

SRC·138 Source
The lamb is lost, and the shepherd doesn't search

The lamb is lost, and the shepherd doesn't search

A lamb goes missing at dusk, high on the far slope. The other herders would fan out with lanterns. Elin just banks the fire, walks down to the pond at the valley's floor, and waits. Before full dark, a small shape trots out of the gloom to the water's edge. She never searched at all. What does she know about this valley?
In bowl country, every downhill road ends at the pond

In bowl country, every downhill road ends at the pond

Her secret is the valley itself: one great bowl. Stand anywhere — the north cliffs, the eastern meadows — and downhill leads, sooner or later, to the same pond. A lost sheep needs no map and no cleverness; it only keeps walking down, and the land itself delivers it to the bottom. But across the ridge, the country keeps a different law…
Across the ridge, downhill is a trap

Across the ridge, downhill is a trap

The badlands: crumpled country, folded into pockets and dead-end hollows. Downhill still feels right underfoot there — and it strands you. Each hollow keeps its own little bottom, and a drifting sheep settles into whichever pit lies nearest. A hundred hollows, a hundred endings; where you stop depends on where you started. So how does Elin tell which law a country obeys?
One taut cord tells which country you're in

One taut cord tells which country you're in

f(λa+(1λ)b)λf(a)+(1λ)f(b)f(\lambda a + (1-\lambda)b) \le \lambda f(a) + (1-\lambda) f(b)
She carries a cord. Stretched taut between any two spots of ground, it tells the truth: in bowl country the land sags beneath the line — the cord never gets buried. In the badlands, some bulge always shoulders over it. The equation below says only that: between any two points, the ground never rises above the straight line joining them. Country that passes the cord test keeps a promise…
The bowl's promise has a name

The bowl's promise has a name

Land that passes the cord test is convex, and convexity is a guarantee: a convex valley has no false bottoms — wherever downhill finally rests you, no lower place exists anywhere. Machines learn this way, stepping downhill on a landscape of error, and on convex ground arrival is promised. The vast crumpled landscapes of modern learning are not convex — yet every morning, the walkers still set out…
🌱 Is this the bottom, or just yours?

🌱 Is this the bottom, or just yours?

At first light Elin sits on the ridge, bowl country on one side, badlands on the other. She thinks of all the downhill walking done in the world with no cord test taken first — habits, careers, quarrels — each life settling into the nearest hollow and calling it the bottom. Once you have settled somewhere, how would you ever know: is this the bottom, or only yours?
tap →swipe ↑ for depthswipe ↓ to exit