Karpathy describes a steep funnel in how people understand AI
Andrej Karpathy, co-founder of OpenAI and former head of AI at Tesla
Understanding of language models narrows like a funnel, Andrej Karpathy argued, from billions who have barely used them to about 5,000 people inside the labs who watch agents work for weeks on mathematics and science.
In the Daily Brief of 12 October 2026
Filed under Large language models, Agents
In a post on X, Andrej Karpathy returned to an observation he first made in April about how unevenly people understand what large language models can do. He now describes that understanding as a steep funnel, with each level seeing a very different technology.
At the wide end, by his estimate, are about 6 billion people who have barely used the models, and 1 to 2 billion who know them only through free chatbots. Below them are some 20 million professionals in mathematics and software who use agents to compress weeks of work into a few prompts. At the narrow end are roughly 5,000 people inside the labs, who watch groups of agents work for weeks at a time on problems in mathematics and science.
The numbers are his rough estimates rather than measurements. Replies to the post turned on how uneven progress has been from one kind of task to another, on the part played by training with rewards that can be checked automatically, and on the gap between public perception of the models and the way they are used at the frontier.