Garry Tan, CEO of Y Combinator, expressed support for allowing American AI labs to use distillation methods on leading AI models, similar to practices observed in Chinese AI labs. Distillation involves extensively querying a model to understand its reasoning and behavior, a technique commonly used to train new AI models.
Tan told CNBC that regulators should avoid restricting this practice and even proposed the idea of an American distillation framework. He envisions smaller U.S.-based open-weight AI labs applying these techniques to frontier AI models to create a more diverse set of open AI options that are not dominated by foreign entities.
This stance contrasts with recent concerns raised by Anthropic, whose CEO Dario Amodei accused Chinese labs of conducting unauthorized distillation attacks using fraudulent credentials. While Tan does not condone illicit activities, he emphasizes that legitimate, authorized distillation should remain unrestricted.
Tan argues that proprietary AI labs should not control how users interact with their models, especially since these labs initially trained their models on vast amounts of publicly available data, often without explicit permission from content owners. He suggests that access to AI intelligence derived from public data should be treated more like a public good rather than being locked behind restrictive terms.
Highlighting the need for a healthy balance, Tan supports both frontier AI labs that push the boundaries of innovation and open-weight models that provide broader access and freedom. He warns against a scenario where a single dominant company monopolizes AI development, which he views as a potential risk to the industry’s future.
Tan’s perspective underscores ongoing debates about AI openness, intellectual property, and the role of regulation in shaping the AI landscape in the United States.