Watch on YouTube ↗

Read the article and sources →

Transcript

Auto-generated YouTube captions. They may contain transcription errors.

  • 0:00 What do the top 5,000 people in AIC that
  • 0:03 the rest of us don't?
  • 0:05 Ask 10 people what AI can do and you'll
  • 0:08 get 10 different answers. Some say it's
  • 0:10 overhyped. Others say it's changing
  • 0:13 everything.
  • 0:14 What if they're both right because
  • 0:16 they're seeing completely different
  • 0:17 things?
  • 0:19 In April, AI researcher Andrej Karpathy
  • 0:22 described this as two groups of people
  • 0:24 talking past each other. Six months
  • 0:26 later, he came back with an update. The
  • 0:29 gap is widening. It isn't two groups
  • 0:32 anymore. It's a funnel.
  • 0:34 Start at the top. By his rough napkin
  • 0:37 math, around 6 billion people, about 3/4
  • 0:40 of humanity, have barely touched an AI
  • 0:43 model at all. For most of the world,
  • 0:45 this revolution hasn't arrived yet.
  • 0:48 One level down are 1 to 2 billion casual
  • 0:51 users. They open a free chatbot now and
  • 0:53 then to look something up or polish a
  • 0:55 paragraph. To them, it's a slightly
  • 0:58 better search engine. Useful, but
  • 1:01 nothing earth-shattering.
  • 1:02 Narrower still are about 20 million
  • 1:05 people, roughly 0.2%,
  • 1:07 who use the most powerful models every
  • 1:09 day for coding and math. They've watched
  • 1:12 an AI agent build or translate an entire
  • 1:15 app from a single prompt. Work that once
  • 1:18 took weeks now takes an afternoon.
  • 1:21 And at the very tip are about 5,000
  • 1:23 people on the entire planet with inside
  • 1:26 access to systems that aren't public
  • 1:28 yet. They're directing swarms of AI
  • 1:30 agents on problems that would take top
  • 1:32 experts years. Karpathy says even human
  • 1:36 review is starting to fall behind.
  • 1:38 Why is the experience so uneven?
  • 1:41 AI improves fastest where answers can be
  • 1:44 checked, where there's plenty of data,
  • 1:46 and where the work is valuable. Coding
  • 1:48 and math hit all three, so the people
  • 1:50 working in those fields see the leap
  • 1:52 first. Karpathy calls this the weirdness
  • 1:55 of the moment. Most of the world hasn't
  • 1:58 used these tools. Most professionals see
  • 2:00 a modest boost and a tiny group is
  • 2:03 watching the curve go vertical, all at
  • 2:06 the same time.
  • 2:08 So, the next time someone tells you what
  • 2:09 AI can or can't do, ask yourself which
  • 2:13 part of the funnel they're standing in
  • 2:15 and where you are.