The gap in shared understanding of LLM capability is widening

Andrej Karpathy argues that understanding of LLM capabilities is increasingly diverging into distinct tiers: 6 billion people have minimal exposure and view LLMs as basic chatbots, about 20 million professionals using frontier-grade systems in math and coding see dramatic productivity gains completing projects in days that once took months, and roughly 5,000 people with internal access to the most advanced systems are witnessing capabilities—like autonomous agents solving complex problems and discovering new science—that far exceed public perception. This widening gap stems from differences in problem scale, domain suitability (where LLMs excel in digital and verifiable domains), and access levels, creating a situation where the general public's understanding of AI remains stalled at an outdated baseline while a small elite experiences exponentially advancing capabilities.