Garry Tan, CEO of Y Combinator, has expressed support for allowing US-based AI labs to employ distillation methods similar to those reportedly used by Chinese AI labs. Distillation involves extensively querying a model to understand its reasoning and behavior, a common practice in AI development for training new models. Tan argues that regulators should not intervene to restrict this practice and that American AI labs should be free to use distillation openly and legitimately.
This stance contrasts with recent concerns raised by Anthropic, whose CEO Dario Amodei has accused Chinese labs of conducting unauthorized "illicit distillation attacks" using deceptive means. While Tan does not condone any illicit activity, he emphasizes that legitimate distillation should be encouraged to foster a diverse and robust AI landscape in the US.
Tan points out that proprietary AI labs previously trained their models on vast amounts of publicly available data without explicit permission, suggesting it would be inconsistent to limit how users interact with these models via API calls. He believes that access to AI intelligence derived from public data should be treated more as a public good rather than being restricted by stringent terms of service.
Highlighting the need for a healthy balance, Tan supports both frontier AI labs pushing the boundaries of innovation and open-weight AI labs that provide broader access. He warns against a scenario where a single dominant company controls AI development, which could stifle competition and innovation. According to Tan, encouraging open distillation practices will help prevent monopolization and ensure a more competitive AI ecosystem in the US.