Karpathy: only 0.2% of people feel what frontier AI can do, and 75% have barely touched an LLM

Andrej Karpathy, an OpenAI co-founder, says the gap in how people understand large language models is widening, and he now describes it as a funnel.

Reposting a thread from April on October 11, Karpathy wrote that the gap in shared understanding of LLM capability is "widening." By Monday morning in Korea the post had about 976,000 views and 16,000 likes.

By his napkin math, about 6 billion people, roughly 75% of the world, have barely come in contact with LLMs. Another 1 to 2 billion, about 20%, are casual users of free chatbots who treat them like a better Google search or a writing aid.

Around 20 million people (0.2%) use frontier-grade models professionally in math and code. They see first-hand that projects that took weeks or months can now be finished by agents with a prompt, he wrote, citing building, copying and decompiling apps. Less than a year ago, he added, he was still typing code by hand.

At the narrow end sit about 5,000 people with internal access to frontier systems, where swarms of agents collaborate for weeks to find zero-days and push the frontier of mathematics, and human review is starting to fall behind.

He blamed the gap on problem size, the jagged capability of models across fields, and access tiers. "The general public has mostly not interacted with these systems," he wrote, and only "a small sliver of professionals" feel "the vertigo of the curve going vertical."