video, optimized

video will become the most programmable, high-bandwidth language computers have ever spoken to us.

our brain is greedy for visuals. a split-second animation packs everything: shape, motion, color, context. all before a sentence even finishes loading in your head. watching is effortless and reading demands effort. that gives video an unfair advantage in the "who grabs attention faster" race. and now we know that "attention is all you need".

veo 3 has crushed video production costs. you type a line of text, wait a few seconds, and a clip with sound pops onto your screen. indistinguishable from reality. this rewrites the race to optimize video content.

by optimize i mean maximize attention.

in tiktok's early days, everything you watched had to be filmed and edited. the recommendation engine then shuffled that fixed library to guess what might keep you watching.

with neural video models, the clip you see doesn't have to exist until the second you hit refresh. the system can generate a brand-new scene, test how long you stay, read the tiny changes in your gaze and pupils, then adjust the next scene to pull you in even harder.

the new loop runs entirely on math. there is no reliance on humans for video anymore. what you get is an infinite scroll that never loosens its grip and an ad engine that knows exactly which pixels make you tap.

karpathy puts it well: when the most absorbing medium on earth becomes fully programmable, we're going to learn what "optimal" looks like. whether we like it or not.

get ready for "optimization".