Tag
Andrej Karpathy argues the gap in shared understanding of LLM capability is widening into a sharp funnel: ~6B people have barely used LLMs, 1-2B are casual free-tier users seeing 'derpy chatbots', ~20M professionals in math&code experience agents completing months of work in prompts, and ~5,000 people internally see frontier systems running swarms of collaborating agents. He attributes the funnel to problem ambition, capability jaggedness (peaks in verifiable domains like math and code), and tiered access.