When it comes to distilling knowledge from cutting-edge AI models, Y Combinator CEO Garry Tan is proposing a hands-off approach, even as some American labs face accusations of illicit practices.

Tan, in an interview with CNBC, suggested that smaller, American AI labs should follow the same distillation techniques used by their Chinese counterparts, arguing that this would create a more robust set of open-weight options for the U.S. market, distinct from Chinese models.

Distillation involves prompting one model to understand another, a legitimate method used to train new models. However, Anthropic, a rival AI lab, has accused Chinese labs of hiding behind stolen credentials to illicitly distill knowledge, a claim they have yet to independently verify.

Tan’s stance is clear: he does not support the use of stolen credentials, but he believes that American labs should have the freedom to ‘come in the front door’ and distill, without government intervention. He argues that controlling how users interact with AI APIs is overly restrictive, and that access to publicly trained intelligence should be considered a public good.

The true worry for Tan is the monopolization of AI power by a single, proprietary provider. He envisions the nightmare scenario where a single company, armed with the best access to capital and AI researchers, dominates the market, creating a monolithic entity that could hinder innovation and access.