Karpathy: Nearly 6 billion people have barely used AI, while only 5,000 have access to internal top-tier models.

Beating AI News Brief: Andrej Karpathy, who currently works on Anthropic’s pre-training team, has again addressed the public perception gap regarding AI capabilities. He argues this gap is widening, evolving from two distinct user groups talking past each other into an increasingly steep funnel. In his rough estimate, roughly 6 billion people (75% of the global population) have almost no exposure to large models. Around 1 billion to 2 billion people only occasionally use free chatbots, treating AI as a search or writing tool. Moving up the funnel, about 20 million people (0.2%) use cutting-edge models in professional fields like programming and mathematics. They have observed that complex projects that once took weeks or even months can now be handled by AI agents. At the very top are roughly 5,000 people (0.00006%), who have internal access to the most advanced AI systems. Karpathy outlines these frontier capabilities as including thousands of AI agents collaborating continuously for weeks, automatically discovering zero-day vulnerabilities, conducting cyber offense and defense at machine speed, and advancing mathematical research. He notes that human ability to review and understand these AI outputs has begun to lag behind. Karpathy attributes the gap to three factors: differing scales of problems users face; faster AI progress in fields like programming and mathematics where results are easily verifiable; and capability and access gaps between free, paid, and internal AI systems. The general public may still view AI as little more than an error-prone chatbot, while a small group of professionals have already started using it to handle large-scale projects.