Karpathy: 6 billion people rarely use AI; only 5,000 access top models

ME AI message: Andrej Karpathy, currently working at Anthropic’s pre-training team, has again addressed the growing gap in public perception of AI capabilities. He believes this gap is widening, evolving from two groups talking past each other into an increasingly steep funnel. According to his rough estimates, approximately 6 billion people (75%) have had almost no exposure to large models. About 1 to 2 billion people occasionally use free chatbots, treating AI as a search or writing tool. Above them, around 20 million people (0.2%) use cutting-edge models in specialized fields such as programming and mathematics. They have already witnessed how complex projects that once took weeks or even months can now be entrusted to agents. At the very top are roughly 5,000 people (0.00006%) who have internal access to the most advanced AI systems. Karpathy describes frontier capabilities including thousands of agents collaborating continuously for weeks, autonomously discovering zero-day vulnerabilities, conducting cyber warfare at machine speed, and advancing mathematical research. He argues that the pace at which humans can review and understand these outcomes is already beginning to lag behind. Karpathy attributes this gap to three factors: the differing scales of problems users face; AI’s faster progress in fields like programming and mathematics where results are easier to verify; and disparities in capability and access between free, paid, and internal systems. While ordinary users may still see AI as a frequently flawed chatbot, a small group of professionals has already begun entrusting it with large-scale projects. (Source: BlockBeats)