Artificial intelligence researcher, OpenAI co-founder, and former Tesla AI director Andrej Karpathy said on X that the public’s shared understanding of large language model capabilities is widening, shifting from two groups talking past each other into a steeper funnel. According to ChainCatcher, he revisited an April post and broke user exposure into tiers based on access level.

Karpathy estimated that about 6 billion people, or roughly 75% of the global population, have barely used large language models. He said another 1 billion to 2 billion people, or about 20%, only occasionally use free ChatGPT-like products and often view them as better search or writing tools, while many non-technical users and professionals outside math and coding see limited utility.

He said about 20 million people, or 0.2%, use frontier models professionally in math and coding, where tasks that once took weeks or months can now be completed by agents under prompts, including building, copying, translating apps, and reverse-engineering applications from binaries. Karpathy added that less than a year ago, he was still writing code manually line by line.

Karpathy also said about 5,000 people, or roughly 0.00006%, have internal access to frontier systems. He described external previews as involving thousands of agents working together for weeks on large software projects, cyber offense and defense, scientific discovery, and the frontier of mathematics, with human review increasingly lagging behind. He said the funnel is driven by problem scale and time horizon, stronger performance in digital and verifiable domains, and the gap between free, paid, and internal access.