Karpathy describes a steepening funnel in who sees what AI can do

Andrej Karpathy returned to an earlier thread to argue that the gap in how people understand AI has sharpened. At the wide end of what he calls a funnel, most people barely use large language models or only see free-tier chatbots.

Further in, a small professional slice working in mathematics and code sees agentic systems complete complex projects in hours. At the narrowest end, he said, an even smaller internal group sees swarms of frontier agents.

He attributed the shape of the funnel to verifiable rewards, the economic focus of labs, and tiered access to the best models. These are his own observations, not measurements.