Karpathy: AI users live in four different worlds

Andrej Karpathy says the gap in how people understand AI is widening. His latest post describes four groups, using rough estimates rather than survey findings.

Barely exposed: around 6 billion people. Most have had little contact with large language models.

Casual users: around 1–2 billion. They use free chatbots occasionally for search, writing or everyday tasks. Many professionals outside mathematics and coding are here too.

Professional users: around 20 million. In mathematics and coding, Karpathy says frontier models can turn projects that once took weeks or months into work agents complete from a prompt.

Internal users: around 5,000. These people have access to frontier systems inside AI companies. He describes thousands of agents working together on software, security, science and mathematics, while human understanding struggles to keep pace.

Karpathy attributes the divide to three things: the ambition and duration of the task, uneven capabilities across domains, and access to free, paid or internal systems.

His point: people can be using different tiers for different tasks and come away with very different impressions of AI.

Read Karpathy’s original post.