Y Combinator CEO Garry Tan is advocating for U.S. open-weight AI labs to freely "distill" frontier models without regulatory restrictions, arguing this would give America more robust open-weight options distinct from Chinese alternatives. Distillation involves extensively prompting another model to learn its reasoning patterns. While Anthropic this week released a report accusing Chinese labs of "illicit distillation attacks" using fraud and stolen credentials, Tan emphasizes he's not proposing theft—he wants American labs to "come in the front door."
Tan frames his argument on two fronts: AI labs shouldn't dictate customer behavior with API calls, and proprietary labs themselves never requested permission when training on copyrighted material. "Controlling what users and customers do with API calls to closed weight models feels constraining," he told TechCrunch, arguing that intelligence trained on public data should function more as a public good than proprietary IP.
Tan, who describes himself as an avid AI user with "cyber psychosis," wants balance between frontier and open-weight labs. He told CNBC that frontier labs are driving progress and deserve fundable business models, while open-weight models provide freedom and access. His ultimate concern is consolidation: "The nightmare scenario, the doomer scenario for AI is that there's just one company" with dominant AI power.