Y Combinator’s Garry Tan wants U.S. open-weight AI labs to ‘distill’ frontier models, too

Garry Tan, the CEO of Y Combinator, has expressed his desire for U.S. AI labs to adopt distillation techniques similar to those used by their Chinese counterparts. In a recent interview with CNBC, he suggested that regulators should refrain from interfering, stating, "I would do nothing." Tan elaborated further in a conversation with TechCrunch, proposing that smaller American open-weight AI labs should leverage training methods on American frontier AI models. This, he believes, would create a diverse array of open-weight options distinct from Chinese innovations. Distillation involves prompting one model to learn from another, a practice that has become a standard, legitimate method for training new AI models. The discussion follows Anthropic's recent report accusing Chinese labs of conducting unauthorized distillation attacks, utilizing hidden identities and fraudulent credentials. In contrast, Tan is not advocating for unethical practices; he emphasizes that American labs should be permitted to operate transparently. He argues that it is excessive for AI labs to impose restrictions on how their models’ outputs are utilized by users. Moreover, he points out that proprietary AI labs did not seek permission when they amassed vast amounts of data for training, often using copyrighted material without consent. "Controlling what users do with API calls to closed-weight models feels limiting," Tan stated, asserting that access to intelligence derived from publicly available data should be considered a public good rather than something restricted by stringent terms of service. Tan, who has passionately engaged with AI to the point of calling himself a cyber-psychosis enthusiast, advocates for a balance between open-weight and frontier AI labs. He believes that while frontier labs are crucial for driving innovation, it is essential to maintain a competitive landscape. "The worst-case scenario for AI is a monopoly by a single company," Tan cautioned. "If one entity monopolizes the best resources and talent, it could lead to detrimental outcomes for the industry and society at large."