Andrej Karpathy says public understanding of LLM capabilities is widening as frontier users see agent-driven breakthroughs

Andrej Karpathy argues that most people still have little or only casual exposure to LLMs, while a tiny group of technical users is seeing much more advanced capabilities firsthand. He says professional users of frontier-grade LLMs can now complete complex coding and math projects far faster, and internal users may be testing agent swarms for software, cybersecurity, science, and mathematics.

The comments highlight a growing AI literacy gap between the public, casual users, professional developers, and insiders with access to the most capable systems. If true, this divergence could shape labor markets, cybersecurity risks, and policy debates, as many decision-makers may be evaluating AI based on outdated or limited experiences.