Distillation as a Strategic Response
Following allegations of illicit distillation attacks by Chinese AI labs, Y Combinator CEO Garry Tan is proposing a counter-strategy for U.S. open-weight AI development. He believes that American labs should engage in similar distillation practices, mirroring the approach taken by Chinese entities. This would foster a more robust open-weight AI landscape within the United States, reducing reliance on a single, potentially dominant, frontier model ecosystem.
Addressing Concerns About Access
Tan’s argument centers on the control that proprietary AI labs exert over access to model insights. He contends that restricting how users and customers interact with closed-weight models through API calls represents an unnecessary constraint. This aligns with his broader view that access to intelligence derived from broad public data should be considered a public good, not a proprietary asset.
Countering Prior Practices
Tan references the historical practice of proprietary AI labs ingesting vast amounts of copyrighted material without explicit permission. He argues that this established a precedent for unrestricted data access, justifying a more open approach to model training and understanding. This stance challenges the current regulatory focus on controlling model outputs and data usage.
Promoting a Diverse Ecosystem
Ultimately, Tan’s goal is to prevent a scenario where a single, powerful proprietary provider dominates the AI landscape. He believes that fostering a diverse ecosystem, including both frontier and open-weight labs, is crucial for innovation and avoiding potential risks associated with concentrated power. Source: https://techcrunch.com/2026/09/11/y-combinators-garry-tan-wants-u-s-open-weight-ai-labs-to-distill-frontier-models-too/
