Karpathy Warns of Growing AI Knowledge Gap as Public Sees Only a Preview
Andrej Karpathy, co-founder of OpenAI and now at Anthropic, says roughly 75 percent of the world's population has barely used large language models, while about 5,000 people have access to internal frontier systems whose agent swarms perform work that would take top experts years. The public, he argues, sees only a weird chatbot.
Andrej Karpathy, a co-founder of OpenAI who now works at Anthropic, has warned that the gap between what the public understands about artificial intelligence and what the technology can actually do is widening rapidly. In his assessment, about 75 percent of the world's population has barely touched large language models, while only a small elite has access to the most advanced internal systems.
Karpathy estimates that roughly 5,000 people have hands-on experience with frontier AI systems whose agent swarms can perform work that would take top human experts years to complete. The external world, he suggests, has seen only the preview of what AI is capable of. For most people, the visible face of the technology remains a quirky chatbot that occasionally produces strange or unreliable answers.
The remarks point to a structural divide that goes beyond ordinary questions of digital access. It is not simply that some people have faster internet or newer devices. The deeper issue is that the most capable AI systems are being developed and tested behind closed doors, available only to a small number of researchers, engineers, and executives. What the public encounters is a heavily filtered, consumer-friendly version of the technology, one that is deliberately limited in scope and capability.
This asymmetry has significant implications for how societies debate AI regulation, safety, and economic impact. If only a tiny fraction of the population has direct knowledge of what advanced AI can do, then public discourse, media coverage, and political decision-making risk being shaped by secondhand impressions rather than direct experience. Karpathy's warning suggests that the gap between the preview and the reality may be far larger than most people assume.
The concentration of access also raises questions about accountability and governance. When a handful of individuals and organizations control the most powerful systems, the rest of the world is left to react to developments it cannot fully observe or evaluate. That dynamic can breed both exaggerated fear and unwarranted complacency, depending on which fragments of information reach the public.
Karpathy's own career reflects the rapid evolution of the field. As a co-founder of OpenAI, he was present at the creation of some of the most influential AI research efforts. His move to Anthropic places him at another leading lab focused on building advanced AI systems with an emphasis on safety and reliability. His perspective is therefore informed by direct exposure to both the public-facing products and the internal capabilities that remain largely hidden.
The 75 percent figure he cites is a reminder that the global adoption of AI is far from uniform. While headlines in technology hubs may suggest that AI is everywhere, the reality for most of the world's population is very different. Many people have never interacted with a large language model, and even those who have may have used only the most basic features. The gap between the frontier and the everyday user is not just a technical detail. It shapes how the technology will be understood, trusted, and integrated into daily life.
For faith communities, cultural institutions, and civic organizations, the warning carries a broader message. Technological change of this magnitude does not unfold evenly, and its moral and social consequences are often felt first by those with the least power to influence its direction. The question of who gets to see the full picture, and who must make do with the preview, is ultimately a question about justice, transparency, and the common good.
Karpathy's comments do not predict a specific outcome, but they do describe a present reality that is easy to overlook. The public conversation about AI is still largely shaped by what is visible, while the most consequential developments remain out of sight. As the technology continues to advance, the distance between the preview and the reality may become one of the defining divides of the coming decade.
6
