Karpathy describes a steep funnel in public understanding of LLM capability
Overview
Andrej Karpathy says public understanding of LLM capability is splitting into a steep funnel, in an X post dated October 11, 2026.
He estimates about 6 billion people have barely used LLMs, 1 to 2 billion use free-tier chatbots casually, roughly 20 million professionals use frontier models for math and code, and about 5,000 people have access to internal frontier systems. He attributes the funnel to three factors: problem ambition and horizon, capability jaggedness concentrated in verifiable digital domains, and access tier.
A repost by Dongxi NLP on the same day cites the same estimates and adds its own text saying only about 5,000 people use the most advanced models internally. The repost describes Karpathy's tweet as dating from April, which conflicts with the October 11, 2026 date of the original post in the first report.
Written by AI from the articles below · updated Oct 11, 5:24 PM ET
Check the sources:
Developments
2 developments
- Oct 11, 4:53 PM ET · 1 articleDongxi NLP reposts Karpathy's funnel of LLM exposure across the worldDongxi NLP: Dongxi NLP reposts Karpathy's funnel of LLM exposure across the world
- Oct 11, 3:12 PM ET · 1 articleKarpathy says LLM capability understanding is splitting into a steep funnelAndrej Karpathy: Karpathy says LLM capability understanding is splitting into a steep funnel
Article timeline
The articles in this story. Times are ET.
Dongxi NLP@dongxi_nlpXDongxi NLP reposts Karpathy's funnel of LLM exposure across the worldAIDongxi NLP reposts Andrej Karpathy's April tweet, which estimates about 6 billion people have barely used LLMs. Karpathy says roughly 20 million professionals see agents complete large math and code projects, while about 5,000 people have internal access to frontier systems. Dongxi NLP's own added text says only about 5,000 people use the most advanced models internally and calls everyone else on the periphery of AI.
Andrej Karpathy@karpathyXKarpathy says LLM capability understanding is splitting into a steep funnelAIAndrej Karpathy says a gap in understanding of LLM capability is widening into a funnel: about 6 billion people have barely used LLMs, around 1-2 billion use free-tier chatbots casually, roughly 20 million professionals use frontier models for math and code, and about 5,000 people have access to internal frontier systems. He attributes the funnel to three factors: problem ambition and horizon, capability jaggedness concentrated in verifiable digital domains, and access tier.
Heat trend
Not enough continuous observations to show a trend yet.