Build Daily

Tinley Park · October 13, 2026
Rob ShocksOct 12, 2026

Karpathy's AI Divide: Which Side Are You On?

What it covers

The creator breaks down Andrej Karpathy's AI capability funnel—splitting the world into four tiers of AI usage—from 6 billion who've barely touched AI to roughly 5,000 lab insiders—and explains why the funnel exists, shares his own realization that most people aren't ambitious enough with these models, and offers a practical way to push a model much harder.

The outline

Tap a timestamp to jump there in the video.

  1. Intro

    The creator introduces the topic of Karpathy's AI divide and frames the question of which side of the gap the viewer is on.

  2. The gap Karpathy noticed

    Karpathy noticed a stark gap in how people actually engage with AI, ranging from no exposure to deep insider access.

  3. The funnel: four layers

    The video lays out Karpathy's napkin-math funnel: roughly 6 billion who've barely used AI, a billion or two on free chatbots, around 20 million on frontier agents like Claude Code and Codex, and about 5,000 insiders inside the labs.

  4. Why the funnel exists

    The creator explains three reasons the AI capability funnel persists and keeps most people at the bottom.

  5. I was building copies

    The creator shares his personal realization that he, like most people, had been building mere copies rather than pushing AI models to their full potential.

  6. The flywheel

    The video describes a flywheel effect that compounds when you push models harder, reinforcing the gap between the layers.

  7. Go and play: what to actually do

    The creator offers a practical, actionable method for pushing a model much harder than most people currently do.

  8. Outro

    The video closes with final thoughts, a call to engage with the course and community, and links to the creator's socials.

The funnel is the product

A solo builder doesn't get to pick which tier of Karpathy's funnel to serve; the product has to collapse the distance between the 6 billion who've barely touched AI and the 5,000 insiders who push models to their limits. The practical takeaway—pushing a model harder than the default—becomes a feature, not a user skill. That's where the shipping work lives.