As reported by The New Stack, Karpathy divides the audience into roughly 6 billion people with little exposure to large language models; 1 to 2 billion casual users of free chatbots; about 20 million professionals using frontier models; and about 5,000 people with internal access to the strongest agent systems. He says the last group can observe more dramatic, agent-scale work, particularly in verifiable domains such as coding and mathematics.
The figures are Karpathy’s estimates, not a published usage study or an audited market measurement. The report does not identify the internal systems, define the access criteria, or provide evidence that the groups are mutually exclusive.
It also does not establish how quickly capabilities are improving, how broadly those systems will be released, or whether exposure to them translates into reliable deployment outcomes.