The Delta of Intelligence is Human Data | Curtis Northcutt (Director of AI Research, Handshake)

High-quality human data serves as the primary driver for advancing frontier AI models, bridging the intelligence gap between successive model generations. Curtis Northcutt, research lead at Handshake and co-founder of Cleanlab, identifies the four fundamental pillars of the AI market—research, hardware, consulting, and human data—and argues that the latter is the most critical for achieving breakthrough performance. By leveraging a vast network of academic experts, Handshake provides the high-fidelity data necessary for post-training and evaluation, moving beyond simple, low-effort labeling tasks. Northcutt introduces "confident learning," a methodology for training models on verified, noise-free information, and highlights the shift toward intelligence augmentation (IA), where AI systems actively tutor and improve human capabilities. Ultimately, the long tail of human expertise remains essential for solving complex, non-verifiable tasks that current generative models cannot master in isolation.