Karpathy: Diverging Views on AI Come Down to How People Use It. He Sorts Users Into Four Groups
His claim is that someone who occasionally uses a free service and someone who applies AI to specialized work are having different experiences. The group sizes are not survey data but Karpathy's rough estimates.
Andrej Karpathy posted on October 11 (local time) that the gap in how people see large language models (LLMs) keeps widening. His argument: assessments diverge sharply depending on which tier of AI you use and what you use it for.
Karpathy broke the experience down into four groups. People who have barely touched an LLM number about 6 billion, he figures, while 1 billion to 2 billion use free services occasionally. He put the number of people using top-tier models for specialized work such as math and coding at roughly 20 million, and those with access to internal systems inside research labs at about 5,000.
These figures are not survey results. They are Karpathy’s rough estimates. Back in April, he had already posted that differing LLM experiences lead to diverging assessments. This time he added his judgment that the gap has grown wider.
Even when people use the same word, “AI,” the level of capability that comes to mind differs between someone who has occasionally tried a free chatbot and someone who uses AI at work every day. Whether the group sizes and experience gaps Karpathy laid out hold up in actual surveys is worth watching.