Karpathy Exposes AI’s ‘Cognitive Fold’: 75% of the World Has Never Touched a Large Model
Andrej Karpathy, a founding member of OpenAI and former director of AI at Tesla, has reignited debate over the true reach of artificial intelligence. In a series of public remarks, Karpathy argued that roughly 75% of the global population—about 6 billion people—has never interacted with a large language model. Even among the remaining 25%, most usage is superficial: casual chats with a chatbot, a few generated images, or autocomplete suggestions. The number of people who have witnessed frontier AI capabilities—autonomous agent swarms conducting cyber offense and defense, or solving advanced mathematical proofs—is closer to 5,000. This gap between mass perception and cutting-edge reality is what Karpathy calls the ‘cognitive fold’ of AI.
The Illusion of Ubiquity
The narrative that AI is everywhere is driven by a highly visible minority. Tech workers, investors, and media amplify use cases that remain alien to most of the world. For crypto, this disconnect matters. Projects touting ‘AI agents’ often cater to the same 25% already saturated with AI tools, while the unbanked and unconnected remain untouched. The real opportunity for decentralized AI—on-chain inference markets, tokenized compute, and agent-to-agent economies—may lie in bridging that gap rather than further serving the converted.
Agent Swarms and the Superintelligence Seed
Karpathy’s observation that top researchers have witnessed agent swarms autonomously attacking and defending networks, and performing mathematical reasoning, suggests that the seeds of superintelligence are already germinating—but behind closed doors. This asymmetry raises concerns about centralization. If only a handful of labs and their close associates can access frontier models, the decentralized ethos of crypto—open access, permissionless innovation—is at risk. Blockchain-based AI networks promise to democratize compute and model access, but they must first solve scalability and verification.
Implications for Crypto and DeAI
- Market education: Crypto AI projects must target the 75% with simple, tangible use cases—not just agent frameworks for developers.
- Verification and trust: As agent swarms grow, on-chain proofs of model behavior and decentralized identity for agents become critical.
- Compute accessibility: Decentralized GPU networks could lower barriers, but only if they can match the performance and reliability of centralized clouds.
Forward Look
The cognitive fold will not close on its own. As AI capabilities accelerate, the risk of a two-tier world—where a tiny elite wields superintelligent agents while billions remain on the outside—grows. Crypto’s role is not guaranteed. It must prove that decentralized infrastructure can deliver AI access that is both trustworthy and inclusive. The next bull market may not be about tokens, but about bridging the fold.

