Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do

Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do

Key Points

• OpenAI co-founder and now Anthropic researcher Andrej Karpathy describes a widening gap in how people understand AI capabilities.
• About 75 percent of the world's population has barely touched large language models. Casual users mostly see them as writing aids or a better search engine.
• Only about 20 million professionals in math and programming use frontier models for work, by Karpathy's estimate. Roughly 5,000 people worldwide have access to internal frontier systems.

Andrej Karpathy, co-founder of OpenAI, describes a rapidly widening gap in how people understand AI capabilities. While the public treats chatbots like a better Google search, internal frontier systems are already tackling tasks that would take top experts years to complete.

Karpathy breaks down the user base of large language models into four levels. About six billion people, roughly 75 percent of the global population, have barely come in contact with LLMs at all. Another one to two billion are casual users of free-tier ChatGPT-like products who treat them "a bit like a better Google search, a writing aid, or etc."

Most people have no idea what AI can already do

Even many professionals outside of math and code fall into this tier. Executives and people in similar roles spend most of their time talking to other people, so "while they understand what is happening intellectually, it is still second-hand and a bit abstract," Karpathy writes.

The third tier covers an estimated 20 million people, about 0.2 percent of the world's population. These are professional users of frontier-grade LLMs in math and code who see firsthand that "large, complex projects that used to take them weeks/months can now be completed by agents with a prompt." Karpathy lists "building apps, copying apps, translating apps, decompiling apps from binaries" as examples.Ad

He adds that this shift happened fast. In October 2024, Karpathy called AI agents mostly marketing that barely worked. Just a few months later, he stopped writing code by hand entirely, going from "typing memorized computer code commands into a code editor character by character" to letting agents handle the work. He then joined Anthropic, despite being an OpenAI co-founder.

A tiny elite sees what the public can't

About 5,000 people, roughly 0.00006 percent of the global population, have access to internal frontier-grade systems by Karpathy's estimate. "The external world has seen the preview," he writes. What this group sees "looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics." These are "things that would have taken top professionals in the industry years of work."

At the same time, "human review and comprehension are starting to fall behind." He points to people still doing "archeology" on the OpenAI-HF incident from many months ago, and notes that mathematicians "may be poring over the 722 manuscripts on frontier mathematics for a while."

Karpathy calls this "the weirdness of the moment." The general public has mostly not interacted with these systems, and "when they have, it looks like a derpy chatbot." Most professionals still see "only a modest uplift." And "a small sliver of professionals are experiencing the vertigo of the curve going vertical." All of it is happening at the same time.Ad

OpenAI CEO Sam Altman made a similar claim in early September at the G20 Innovation Ministerial, saying a current AI coding tool could achieve in "17 minutes" what a startup had previously built over three months. "The world has not caught up with what this is going to mean," Altman said.

Karpathy says LLM capability is currently peaking in domains that are "digital, verifiable and economically valuable." That's because it "emerges from reinforcement learning on verifiable rewards," and companies direct their training resources toward the domains with the highest revenue potential. Right now, that means coding and math.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.